modelId
string | author
string | last_modified
unknown | downloads
int64 | likes
int64 | library_name
string | tags
sequence | pipeline_tag
string | createdAt
unknown | card
string |
|---|---|---|---|---|---|---|---|---|---|
buelfhood/conplag1_codebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_pfeiffer | buelfhood | "2025-05-06T23:36:07Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T23:36:04Z" | ---
tags:
- roberta
- adapter-transformers
---
# Adapter `buelfhood/conplag1_codebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_pfeiffer` for microsoft/codebert-base
An [adapter](https://adapterhub.ml) for the `microsoft/codebert-base` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("microsoft/codebert-base")
adapter_name = model.load_adapter("buelfhood/conplag1_codebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_pfeiffer", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF | mradermacher | "2025-05-06T23:32:21Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"en",
"base_model:andrewzh/Absolute_Zero_Reasoner-Coder-3b",
"base_model:quantized:andrewzh/Absolute_Zero_Reasoner-Coder-3b",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-06T21:27:04Z" | ---
base_model: andrewzh/Absolute_Zero_Reasoner-Coder-3b
language:
- en
library_name: transformers
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/andrewzh/Absolute_Zero_Reasoner-Coder-3b
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q2_K.gguf) | Q2_K | 1.4 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q3_K_S.gguf) | Q3_K_S | 1.6 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q3_K_M.gguf) | Q3_K_M | 1.7 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q3_K_L.gguf) | Q3_K_L | 1.8 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.IQ4_XS.gguf) | IQ4_XS | 1.9 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q4_K_S.gguf) | Q4_K_S | 1.9 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q4_K_M.gguf) | Q4_K_M | 2.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q5_K_S.gguf) | Q5_K_S | 2.3 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q5_K_M.gguf) | Q5_K_M | 2.3 | |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q6_K.gguf) | Q6_K | 2.6 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.Q8_0.gguf) | Q8_0 | 3.4 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Absolute_Zero_Reasoner-Coder-3b-GGUF/resolve/main/Absolute_Zero_Reasoner-Coder-3b.f16.gguf) | f16 | 6.3 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
rayonlabs/hf-autotrain-2025-05-05-f815c2b3 | rayonlabs | "2025-05-06T23:28:31Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"autotrain",
"text-generation-inference",
"peft",
"conversational",
"dataset:rayonlabs/autotrain-data-hf-autotrain-2025-05-05-f815c2b3",
"base_model:unsloth/Qwen2-7B-Instruct",
"base_model:finetune:unsloth/Qwen2-7B-Instruct",
"license:other",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-05T21:11:08Z" | ---
tags:
- autotrain
- text-generation-inference
- text-generation
- peft
library_name: transformers
base_model: unsloth/Qwen2-7B-Instruct
widget:
- messages:
- role: user
content: What is your favorite condiment?
license: other
datasets:
- rayonlabs/autotrain-data-hf-autotrain-2025-05-05-f815c2b3
---
# Model Trained Using AutoTrain
This model was trained using AutoTrain. For more information, please visit [AutoTrain](https://hf.co/docs/autotrain).
# Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_path = "PATH_TO_THIS_REPO"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
model_path,
device_map="auto",
torch_dtype='auto'
).eval()
# Prompt content: "hi"
messages = [
{"role": "user", "content": "hi"}
]
input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(input_ids.to('cuda'))
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
# Model response: "Hello! How can I assist you today?"
print(response)
``` |
kazuyamaa/Qwen2.5-3B-Instruct-GRPO-v001 | kazuyamaa | "2025-05-06T23:25:29Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"text-generation-inference",
"unsloth",
"trl",
"grpo",
"conversational",
"en",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"8-bit",
"region:us"
] | text-generation | "2025-05-06T16:33:31Z" | ---
base_model: unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
- grpo
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** kazuyamaa
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
launchpd3/gensyn-checkpoints-quick_eager_hippo | launchpd3 | "2025-05-06T23:23:46Z" | 2 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"generated_from_trainer",
"rl-swarm",
"grpo",
"gensyn",
"I am quick eager hippo",
"unsloth",
"trl",
"conversational",
"arxiv:2402.03300",
"base_model:Gensyn/Qwen2.5-1.5B-Instruct",
"base_model:finetune:Gensyn/Qwen2.5-1.5B-Instruct",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-04-23T05:34:11Z" | ---
base_model: Gensyn/Qwen2.5-1.5B-Instruct
library_name: transformers
model_name: gensyn-checkpoints-quick_eager_hippo
tags:
- generated_from_trainer
- rl-swarm
- grpo
- gensyn
- I am quick eager hippo
- unsloth
- trl
licence: license
---
# Model Card for gensyn-checkpoints-quick_eager_hippo
This model is a fine-tuned version of [Gensyn/Qwen2.5-1.5B-Instruct](https://huggingface.co/Gensyn/Qwen2.5-1.5B-Instruct).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="launchpd3/gensyn-checkpoints-quick_eager_hippo", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
### Framework versions
- TRL: 0.15.2
- Transformers: 4.51.3
- Pytorch: 2.6.0
- Datasets: 3.5.0
- Tokenizers: 0.21.1
## Citations
Cite GRPO as:
```bibtex
@article{zhihong2024deepseekmath,
title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
year = 2024,
eprint = {arXiv:2402.03300},
}
```
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
fixie-ai/ultravox-v0_5-llama-3_2-1b | fixie-ai | "2025-05-06T23:23:31Z" | 174,953 | 22 | transformers | [
"transformers",
"safetensors",
"ultravox",
"feature-extraction",
"audio-text-to-text",
"custom_code",
"ar",
"be",
"bg",
"bn",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",
"fa",
"fi",
"fr",
"gl",
"hi",
"hu",
"it",
"ja",
"ka",
"lt",
"lv",
"mk",
"mr",
"nl",
"pl",
"pt",
"ro",
"ru",
"sk",
"sl",
"sr",
"sv",
"sw",
"ta",
"th",
"tr",
"uk",
"ur",
"vi",
"zh",
"license:mit",
"region:us"
] | audio-text-to-text | "2025-02-06T22:48:54Z" | ---
language:
- ar
- be
- bg
- bn
- cs
- cy
- da
- de
- el
- en
- es
- et
- fa
- fi
- fr
- gl
- hi
- hu
- it
- ja
- ka
- lt
- lv
- mk
- mr
- nl
- pl
- pt
- ro
- ru
- sk
- sl
- sr
- sv
- sw
- ta
- th
- tr
- uk
- ur
- vi
- zh
library_name: transformers
license: mit
metrics:
- bleu
pipeline_tag: audio-text-to-text
---
# Model Card for Ultravox
Ultravox is a multimodal Speech LLM built around a pretrained [Llama3.2-1B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-1B) and [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) backbone.
See https://ultravox.ai for the GitHub repo and more information.
## Model Details
### Model Description
Ultravox is a multimodal model that can consume both speech and text as input (e.g., a text system prompt and voice user message).
The input to the model is given as a text prompt with a special `<|audio|>` pseudo-token, and the model processor will replace this magic token with embeddings derived from the input audio.
Using the merged embeddings as input, the model will then generate output text as usual.
In a future revision of Ultravox, we plan to expand the token vocabulary to support generation of semantic and acoustic audio tokens, which can then be fed to a vocoder to produce voice output.
No preference tuning has been applied to this revision of the model.
- **Developed by:** Fixie.ai
- **License:** MIT
### Model Sources
- **Repository:** https://ultravox.ai
- **Demo:** See repo
## Usage
Think of the model as an LLM that can also hear and understand speech. As such, it can be used as a voice agent, and also to do speech-to-speech translation, analysis of spoken audio, etc.
To use the model, try the following:
```python
# pip install transformers peft librosa
import transformers
import numpy as np
import librosa
pipe = transformers.pipeline(model='fixie-ai/ultravox-v0_5-llama-3_2-1b', trust_remote_code=True)
path = "<path-to-input-audio>" # TODO: pass the audio here
audio, sr = librosa.load(path, sr=16000)
turns = [
{
"role": "system",
"content": "You are a friendly and helpful character. You love to answer questions for people."
},
]
pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=30)
```
## Training Details
The model uses a pre-trained [Llama3.2-1B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.2-1B) backbone as well as the encoder part of [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo).
The multi-modal adapter is trained, the Whisper encoder is fine-tuned, while the Llama model is kept frozen.
We use a knowledge-distillation loss where Ultravox is trying to match the logits of the text-based Llama backbone.
### Training Data
The training dataset is a mix of ASR datasets, extended with continuations generated by Llama 3.1 8B, and speech translation datasets, which yield a modest improvement in translation evaluations.
### Training Procedure
Supervised speech instruction finetuning via knowledge-distillation. For more info, see [training code in Ultravox repo](https://github.com/fixie-ai/ultravox/blob/main/ultravox/training/train.py).
#### Training Hyperparameters
- **Training regime:** BF16 mixed precision training
- **Hardward used:** 8x H100 GPUs
#### Speeds, Sizes, Times
Check out the audio tab on [TheFastest.ai](https://thefastest.ai/?m=audio) for daily benchmarks and a comparison with other existing models.
## Evaluation
| | **Ultravox 0.5 1b**| Ultravox 0.5 8B | Ultravox 0.5 70B |
| --- | ---: | ---: | ---: |
| **covost2 en_ar** | 1.55 | 12.99 | 20.21 |
| **covost2 en_ca** | 8.06 | 31.54 | 40.01 |
| **covost2 en_de** | 14.21 | 28.70 | 34.53 |
| **covost2 es_en** | 24.97 | 40.19 | 43.29 |
| **covost2 ru_en** | 24.12 | 42.13 | 48.99 |
| **covost2 zh_en** | 4.76 | 17.22 | 21.37 |
| **big bench audio**| 39.14 | 66.54 | 82.70 |
|
fixie-ai/ultravox-v0_4_1-llama-3_3-70b | fixie-ai | "2025-05-06T23:23:01Z" | 39 | 10 | transformers | [
"transformers",
"safetensors",
"ultravox",
"feature-extraction",
"audio-text-to-text",
"custom_code",
"ar",
"de",
"en",
"es",
"fr",
"hi",
"it",
"ja",
"nl",
"pt",
"ru",
"sv",
"tr",
"uk",
"zh",
"dataset:fixie-ai/librispeech_asr",
"dataset:fixie-ai/common_voice_17_0",
"dataset:fixie-ai/peoples_speech",
"dataset:fixie-ai/gigaspeech",
"dataset:fixie-ai/multilingual_librispeech",
"dataset:fixie-ai/wenetspeech",
"dataset:fixie-ai/covost2",
"license:mit",
"region:us"
] | audio-text-to-text | "2024-12-16T16:14:37Z" | ---
language:
- ar
- de
- en
- es
- fr
- hi
- it
- ja
- nl
- pt
- ru
- sv
- tr
- uk
- zh
license: mit
library_name: transformers
datasets:
- fixie-ai/librispeech_asr
- fixie-ai/common_voice_17_0
- fixie-ai/peoples_speech
- fixie-ai/gigaspeech
- fixie-ai/multilingual_librispeech
- fixie-ai/wenetspeech
- fixie-ai/covost2
metrics:
- bleu
pipeline_tag: audio-text-to-text
---
# Model Card for Ultravox
Ultravox is a multimodal Speech LLM built around a pretrained [Llama3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) and [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) backbone.
See https://ultravox.ai for the GitHub repo and more information.
## Model Details
### Model Description
Ultravox is a multimodal model that can consume both speech and text as input (e.g., a text system prompt and voice user message).
The input to the model is given as a text prompt with a special `<|audio|>` pseudo-token, and the model processor will replace this magic token with embeddings derived from the input audio.
Using the merged embeddings as input, the model will then generate output text as usual.
In a future revision of Ultravox, we plan to expand the token vocabulary to support generation of semantic and acoustic audio tokens, which can then be fed to a vocoder to produce voice output.
No preference tuning has been applied to this revision of the model.
- **Developed by:** Fixie.ai
- **License:** MIT
### Model Sources
- **Repository:** https://ultravox.ai
- **Demo:** See repo
## Usage
Think of the model as an LLM that can also hear and understand speech. As such, it can be used as a voice agent, and also to do speech-to-speech translation, analysis of spoken audio, etc.
To use the model, try the following:
```python
# pip install transformers peft librosa
import transformers
import numpy as np
import librosa
pipe = transformers.pipeline(model='fixie-ai/ultravox-v0_4_1-llama-3_1-70b', trust_remote_code=True)
path = "<path-to-input-audio>" # TODO: pass the audio here
audio, sr = librosa.load(path, sr=16000)
turns = [
{
"role": "system",
"content": "You are a friendly and helpful character. You love to answer questions for people."
},
]
pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=30)
```
## Training Details
The model uses a pre-trained [Llama3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) backbone as well as the encoder part of [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo).
Only the multi-modal adapter is trained, while Whisper encoder and Llama are kept frozen.
We use a knowledge-distillation loss where Ultravox is trying to match the logits of the text-based Llama backbone.
### Training Data
The training dataset is a mix of ASR datasets, extended with continuations generated by Llama 3.1 8B, and speech translation datasets, which yield a modest improvement in translation evaluations.
### Training Procedure
Supervised speech instruction finetuning via knowledge-distillation. For more info, see [training code in Ultravox repo](https://github.com/fixie-ai/ultravox/blob/main/ultravox/training/train.py).
#### Training Hyperparameters
- **Training regime:** BF16 mixed precision training
- **Hardward used:** 8x H100 GPUs
#### Speeds, Sizes, Times
The current version of Ultravox, when invoked with audio content, has a time-to-first-token (TTFT) of approximately 150ms, and a tokens-per-second rate of ~50-100 when using an A100-40GB GPU, all using a Llama 3.3 70B backbone.
Check out the audio tab on [TheFastest.ai](https://thefastest.ai/?m=audio) for daily benchmarks and a comparison with other existing models.
## Evaluation
| | Ultravox 0.4 70B | Ultravox 0.4.1 70B |
| --- | ---: | ---: |
| **en_ar** | 14.97 | 19.64 |
| **en_de** | 30.30 | 32.47 |
| **es_en** | 39.55 | 40.76 |
| **ru_en** | 44.16 | 45.07 |
| **en_ca** | 35.02 | 37.58 |
| **zh_en** | 12.16 | 17.98 | |
fixie-ai/ultravox-v0_4_1-mistral-nemo | fixie-ai | "2025-05-06T23:22:23Z" | 1,157 | 25 | transformers | [
"transformers",
"safetensors",
"ultravox",
"feature-extraction",
"audio-text-to-text",
"custom_code",
"ar",
"de",
"en",
"es",
"fr",
"hi",
"it",
"ja",
"nl",
"pt",
"ru",
"sv",
"tr",
"uk",
"zh",
"dataset:fixie-ai/librispeech_asr",
"dataset:fixie-ai/common_voice_17_0",
"dataset:fixie-ai/peoples_speech",
"dataset:fixie-ai/gigaspeech",
"dataset:fixie-ai/multilingual_librispeech",
"dataset:fixie-ai/wenetspeech",
"dataset:fixie-ai/covost2",
"license:mit",
"region:us"
] | audio-text-to-text | "2024-11-07T22:46:20Z" | ---
language:
- ar
- de
- en
- es
- fr
- hi
- it
- ja
- nl
- pt
- ru
- sv
- tr
- uk
- zh
license: mit
library_name: transformers
datasets:
- fixie-ai/librispeech_asr
- fixie-ai/common_voice_17_0
- fixie-ai/peoples_speech
- fixie-ai/gigaspeech
- fixie-ai/multilingual_librispeech
- fixie-ai/wenetspeech
- fixie-ai/covost2
metrics:
- bleu
pipeline_tag: audio-text-to-text
---
# Model Card for Ultravox
Ultravox is a multimodal Speech LLM built around a pretrained [Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407) and [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) backbone.
See https://ultravox.ai for the GitHub repo and more information.
## Model Details
### Model Description
Ultravox is a multimodal model that can consume both speech and text as input (e.g., a text system prompt and voice user message).
The input to the model is given as a text prompt with a special `<|audio|>` pseudo-token, and the model processor will replace this magic token with embeddings derived from the input audio.
Using the merged embeddings as input, the model will then generate output text as usual.
In a future revision of Ultravox, we plan to expand the token vocabulary to support generation of semantic and acoustic audio tokens, which can then be fed to a vocoder to produce voice output.
No preference tuning has been applied to this revision of the model.
- **Developed by:** Fixie.ai
- **License:** MIT
### Model Sources
- **Repository:** https://ultravox.ai
- **Demo:** See repo
## Usage
Think of the model as an LLM that can also hear and understand speech. As such, it can be used as a voice agent, and also to do speech-to-speech translation, analysis of spoken audio, etc.
To use the model, try the following:
```python
# pip install transformers peft librosa
import transformers
import numpy as np
import librosa
pipe = transformers.pipeline(model='fixie-ai/ultravox-v0_4_1-mistral-nemo', trust_remote_code=True)
path = "<path-to-input-audio>" # TODO: pass the audio here
audio, sr = librosa.load(path, sr=16000)
turns = [
{
"role": "system",
"content": "You are a friendly and helpful character. You love to answer questions for people."
},
]
pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=30)
```
## Training Details
The model uses a pre-trained [Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407) backbone as well as the encoder part of [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo).
Only the multi-modal adapter is trained, while Whisper encoder and Mistral are kept frozen.
We use a knowledge-distillation loss where Ultravox is trying to match the logits of the text-based Mistral backbone.
### Training Data
The training dataset is a mix of ASR datasets, extended with continuations generated by Mistral Nemo, and speech translation datasets, which yield a modest improvement in translation evaluations.
### Training Procedure
Supervised speech instruction finetuning via knowledge-distillation. For more info, see [training code in Ultravox repo](https://github.com/fixie-ai/ultravox/blob/main/ultravox/training/train.py).
#### Training Hyperparameters
- **Training regime:** BF16 mixed precision training
- **Hardward used:** 8x H100 GPUs
#### Speeds, Sizes, Times
The current version of Ultravox, when invoked with audio content, has a time-to-first-token (TTFT) of approximately 150ms, and a tokens-per-second rate of ~50-100 when using an A100-40GB GPU, all using a Mistral Nemo backbone.
Check out the audio tab on [TheFastest.ai](https://thefastest.ai/?m=audio) for daily benchmarks and a comparison with other existing models.
## Evaluation
| | Ultravox 0.4.1 Mistral Nemo |
| --- | ---: |
| **en_ar** | 10.36 |
| **en_de** | 28.39 |
| **es_en** | 37.49 |
| **ru_en** | 41.64 |
| **en_ca** | 26.85 |
| **zh_en** | 12.65 | |
fixie-ai/ultravox-v0_4_1-llama-3_1-8b | fixie-ai | "2025-05-06T23:22:10Z" | 662 | 98 | transformers | [
"transformers",
"safetensors",
"ultravox",
"feature-extraction",
"audio-text-to-text",
"custom_code",
"ar",
"de",
"en",
"es",
"fr",
"hi",
"it",
"ja",
"nl",
"pt",
"ru",
"sv",
"tr",
"uk",
"zh",
"dataset:fixie-ai/librispeech_asr",
"dataset:fixie-ai/common_voice_17_0",
"dataset:fixie-ai/peoples_speech",
"dataset:fixie-ai/gigaspeech",
"dataset:fixie-ai/multilingual_librispeech",
"dataset:fixie-ai/wenetspeech",
"dataset:fixie-ai/covost2",
"license:mit",
"region:us"
] | audio-text-to-text | "2024-11-05T03:24:47Z" | ---
datasets:
- fixie-ai/librispeech_asr
- fixie-ai/common_voice_17_0
- fixie-ai/peoples_speech
- fixie-ai/gigaspeech
- fixie-ai/multilingual_librispeech
- fixie-ai/wenetspeech
- fixie-ai/covost2
language:
- ar
- de
- en
- es
- fr
- hi
- it
- ja
- nl
- pt
- ru
- sv
- tr
- uk
- zh
library_name: transformers
license: mit
metrics:
- bleu
pipeline_tag: audio-text-to-text
---
# Model Card for Ultravox
Ultravox is a multimodal Speech LLM built around a pretrained [Llama3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B) and [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) backbone.
See https://ultravox.ai for the GitHub repo and more information.
## Model Details
### Model Description
Ultravox is a multimodal model that can consume both speech and text as input (e.g., a text system prompt and voice user message).
The input to the model is given as a text prompt with a special `<|audio|>` pseudo-token, and the model processor will replace this magic token with embeddings derived from the input audio.
Using the merged embeddings as input, the model will then generate output text as usual.
In a future revision of Ultravox, we plan to expand the token vocabulary to support generation of semantic and acoustic audio tokens, which can then be fed to a vocoder to produce voice output.
No preference tuning has been applied to this revision of the model.
- **Developed by:** Fixie.ai
- **License:** MIT
### Model Sources
- **Repository:** https://ultravox.ai
- **Demo:** See repo
## Usage
Think of the model as an LLM that can also hear and understand speech. As such, it can be used as a voice agent, and also to do speech-to-speech translation, analysis of spoken audio, etc.
To use the model, try the following:
```python
# pip install transformers peft librosa
import transformers
import numpy as np
import librosa
pipe = transformers.pipeline(model='fixie-ai/ultravox-v0_4_1-llama-3_1-8b', trust_remote_code=True)
path = "<path-to-input-audio>" # TODO: pass the audio here
audio, sr = librosa.load(path, sr=16000)
turns = [
{
"role": "system",
"content": "You are a friendly and helpful character. You love to answer questions for people."
},
]
pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=30)
```
## Training Details
The model uses a pre-trained [Llama3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B) backbone as well as the encoder part of [whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo).
Only the multi-modal adapter is trained, while Whisper encoder and Llama are kept frozen.
We use a knowledge-distillation loss where Ultravox is trying to match the logits of the text-based Llama backbone.
### Training Data
The training dataset is a mix of ASR datasets, extended with continuations generated by Llama 3.1 8B, and speech translation datasets, which yield a modest improvement in translation evaluations.
### Training Procedure
Supervised speech instruction finetuning via knowledge-distillation. For more info, see [training code in Ultravox repo](https://github.com/fixie-ai/ultravox/blob/main/ultravox/training/train.py).
#### Training Hyperparameters
- **Training regime:** BF16 mixed precision training
- **Hardward used:** 8x H100 GPUs
#### Speeds, Sizes, Times
The current version of Ultravox, when invoked with audio content, has a time-to-first-token (TTFT) of approximately 150ms, and a tokens-per-second rate of ~50-100 when using an A100-40GB GPU, all using a Llama 3.1 8B backbone.
Check out the audio tab on [TheFastest.ai](https://thefastest.ai/?m=audio) for daily benchmarks and a comparison with other existing models.
## Evaluation
| | Ultravox 0.4 8B | **Ultravox 0.4.1 8B** |
| --- | ---: | ---: |
| **en_ar** | 11.17 | 12.28 |
| **en_de** | 25.47 | 27.13 |
| **es_en** | 37.11 | 39.16 |
| **ru_en** | 38.96 | 39.65 |
| **en_ca** | 27.46 | 29.94 |
| **zh_en** | 10.08 | 14.55 | |
fixie-ai/ultravox-v0_4-ToolACE-8B | fixie-ai | "2025-05-06T23:22:00Z" | 1,006 | 1 | transformers | [
"transformers",
"safetensors",
"ultravox",
"feature-extraction",
"audio-text-to-text",
"custom_code",
"arxiv:1910.09700",
"region:us"
] | audio-text-to-text | "2024-10-23T18:15:30Z" | ---
library_name: transformers
pipeline_tag: audio-text-to-text
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
dslighfdsl/Llama-3.1-8B-Instruct-SFT-CoT-short-full-3 | dslighfdsl | "2025-05-06T23:14:42Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"llama",
"text-generation",
"generated_from_trainer",
"open-r1",
"trl",
"sft",
"conversational",
"dataset:sciworld",
"base_model:meta-llama/Llama-3.1-8B-Instruct",
"base_model:finetune:meta-llama/Llama-3.1-8B-Instruct",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-06T21:52:49Z" | ---
base_model: meta-llama/Llama-3.1-8B-Instruct
datasets: sciworld
library_name: transformers
model_name: Llama-3.1-8B-Instruct-SFT-CoT-short-full-3
tags:
- generated_from_trainer
- open-r1
- trl
- sft
licence: license
---
# Model Card for Llama-3.1-8B-Instruct-SFT-CoT-short-full-3
This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on the [sciworld](https://huggingface.co/datasets/sciworld) dataset.
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="dslighfdsl/Llama-3.1-8B-Instruct-SFT-CoT-short-full-3", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/pengliangji2023-carnegie-mellon-university/huggingface/runs/ohdllwpn)
This model was trained with SFT.
### Framework versions
- TRL: 0.15.2
- Transformers: 4.50.0.dev0
- Pytorch: 2.5.1
- Datasets: 3.3.2
- Tokenizers: 0.21.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
Geumokmiddleschool/GeumokGPT-Lite | Geumokmiddleschool | "2025-05-06T23:11:52Z" | 0 | 0 | null | [
"safetensors",
"gemma3_text",
"text-generation-inference",
"ko",
"license:mit",
"region:us"
] | null | "2025-05-06T16:53:18Z" | ---
tags:
- text-generation-inference
license: mit
language:
- ko
---
# Uploaded finetuned model
- **Developed by:** Geumokmiddleschool
- **License:** apache-2.0
Geumok middle school's official Large Language Model
금옥중학교 2학년 1반의 공식 언어모델 GeumokGPT입니다. |
mradermacher/OceanGPT-coder-7B-i1-GGUF | mradermacher | "2025-05-06T23:09:34Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"en",
"base_model:zjunlp/OceanGPT-coder-7B",
"base_model:quantized:zjunlp/OceanGPT-coder-7B",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-05-06T22:13:07Z" | ---
base_model: zjunlp/OceanGPT-coder-7B
language:
- en
library_name: transformers
license: mit
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/zjunlp/OceanGPT-coder-7B
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/OceanGPT-coder-7B-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ1_S.gguf) | i1-IQ1_S | 2.0 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ1_M.gguf) | i1-IQ1_M | 2.1 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.4 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.6 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ2_S.gguf) | i1-IQ2_S | 2.7 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ2_M.gguf) | i1-IQ2_M | 2.9 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q2_K_S.gguf) | i1-Q2_K_S | 2.9 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q2_K.gguf) | i1-Q2_K | 3.1 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 3.2 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ3_XS.gguf) | i1-IQ3_XS | 3.4 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.6 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ3_S.gguf) | i1-IQ3_S | 3.6 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ3_M.gguf) | i1-IQ3_M | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q3_K_M.gguf) | i1-Q3_K_M | 3.9 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q3_K_L.gguf) | i1-Q3_K_L | 4.2 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ4_XS.gguf) | i1-IQ4_XS | 4.3 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-IQ4_NL.gguf) | i1-IQ4_NL | 4.5 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q4_0.gguf) | i1-Q4_0 | 4.5 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.6 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q4_K_M.gguf) | i1-Q4_K_M | 4.8 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q4_1.gguf) | i1-Q4_1 | 5.0 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q5_K_S.gguf) | i1-Q5_K_S | 5.4 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q5_K_M.gguf) | i1-Q5_K_M | 5.5 | |
| [GGUF](https://huggingface.co/mradermacher/OceanGPT-coder-7B-i1-GGUF/resolve/main/OceanGPT-coder-7B.i1-Q6_K.gguf) | i1-Q6_K | 6.4 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
|
Curative/distilbert-ner | Curative | "2025-05-06T23:07:36Z" | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"distilbert",
"token-classification",
"generated_from_trainer",
"dataset:conll2003",
"base_model:distilbert/distilbert-base-cased",
"base_model:finetune:distilbert/distilbert-base-cased",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | "2025-05-06T22:49:40Z" | ---
library_name: transformers
license: apache-2.0
base_model: distilbert-base-cased
tags:
- generated_from_trainer
datasets:
- conll2003
metrics:
- precision
- recall
- f1
- accuracy
model-index:
- name: distilbert-ner
results:
- task:
name: Token Classification
type: token-classification
dataset:
name: conll2003
type: conll2003
config: conll2003
split: validation
args: conll2003
metrics:
- name: Precision
type: precision
value: 0.9300033411293017
- name: Recall
type: recall
value: 0.9368899360484685
- name: F1
type: f1
value: 0.9334339369550636
- name: Accuracy
type: accuracy
value: 0.9889023013122542
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-ner
This model is a fine-tuned version of [distilbert-base-cased](https://huggingface.co/distilbert-base-cased) on the conll2003 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0454
- Precision: 0.9300
- Recall: 0.9369
- F1: 0.9334
- Accuracy: 0.9889
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
| 0.182 | 1.0 | 878 | 0.0564 | 0.9023 | 0.9111 | 0.9067 | 0.9843 |
| 0.038 | 2.0 | 1756 | 0.0504 | 0.9253 | 0.9298 | 0.9276 | 0.9876 |
| 0.0208 | 3.0 | 2634 | 0.0454 | 0.9300 | 0.9369 | 0.9334 | 0.9889 |
### Framework versions
- Transformers 4.51.3
- Pytorch 2.6.0+cu124
- Datasets 3.5.1
- Tokenizers 0.21.1
|
henryhe0123/pc-agent-test-41 | henryhe0123 | "2025-05-06T23:07:31Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2_5_vl",
"image-text-to-text",
"llama-factory",
"full",
"generated_from_trainer",
"conversational",
"base_model:henryhe0123/pc-agent-test-41",
"base_model:finetune:henryhe0123/pc-agent-test-41",
"license:other",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | image-text-to-text | "2025-05-06T18:00:28Z" | ---
library_name: transformers
license: other
base_model: henryhe0123/pc-agent-test-41
tags:
- llama-factory
- full
- generated_from_trainer
model-index:
- name: Qwen2.5-VL-72B-sft-41
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# Qwen2.5-VL-72B-sft-41
This model is a fine-tuned version of [/inspire/hdd/global_user/liupengfei-24025/yhhe/model/Qwen2.5-VL-72B-Instruct](https://huggingface.co//inspire/hdd/global_user/liupengfei-24025/yhhe/model/Qwen2.5-VL-72B-Instruct) on the pcagent41 dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-06
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 32
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 256
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.05
- num_epochs: 2
### Training results
### Framework versions
- Transformers 4.49.0.dev0
- Pytorch 2.6.0+cu124
- Datasets 3.3.2
- Tokenizers 0.21.0
|
GIGAParviz/Parviz_Mind_Reasoning | GIGAParviz | "2025-05-06T23:06:14Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"qwen3",
"trl",
"en",
"base_model:unsloth/Qwen3-4B-unsloth-bnb-4bit",
"base_model:finetune:unsloth/Qwen3-4B-unsloth-bnb-4bit",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T23:05:30Z" | ---
base_model: unsloth/Qwen3-4B-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** GIGAParviz
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Qwen3-4B-unsloth-bnb-4bit
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
Mungert/mOrpheus_3B-1Base_early_preview-v1-25000-GGUF | Mungert | "2025-05-06T23:01:25Z" | 902 | 0 | null | [
"gguf",
"unsloth",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-27T09:31:57Z" | ---
license: cc-by-nc-4.0
tags:
- unsloth
---
# <span style="color: #7FFF7F;">mOrpheus_3B-1Base_early_preview-v1-25000 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`e291450`](https://github.com/ggerganov/llama.cpp/commit/e291450b7602d7a36239e4ceeece37625f838373).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `mOrpheus_3B-1Base_early_preview-v1-25000-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `mOrpheus_3B-1Base_early_preview-v1-25000-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `mOrpheus_3B-1Base_early_preview-v1-25000-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `mOrpheus_3B-1Base_early_preview-v1-25000-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `mOrpheus_3B-1Base_early_preview-v1-25000-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `mOrpheus_3B-1Base_early_preview-v1-25000-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `mOrpheus_3B-1Base_early_preview-v1-25000-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `mOrpheus_3B-1Base_early_preview-v1-25000-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `mOrpheus_3B-1Base_early_preview-v1-25000-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `mOrpheus_3B-1Base_early_preview-v1-25000-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `mOrpheus_3B-1Base_early_preview-v1-25000-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# mOrpheus_3B-1Base_early_preview (NSFW TTS)
A finetuned Orpheus text‑to‑speech model trained on adult data for more expressive sounds:
`<laugh>, <chuckle>, <sigh>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>`
New in this model: `<moans>, <panting>, <grunting>, <gagging sounds>, <chokeing>, <kissing noises>`
**Speaker name:** `baddy`
**Framework:** Safetensors (LLaMA)
**Status:** Early preview; training still underway
---
## 🔗 Links
- Model files & versions: [xet](<your-file-hosting-link>)
- Discussion & bug reports: [Discord server](https://discord.gg/RUs3uzBdW3)
- Original author: [MrDragonFox](https://huggingface.co/MrDragonFox)
---
## 🚀 Usage (Example)
1. Load the `*.GGUF` file into LMStudio.
2. ```bash
pip install RealtimeTTS[orpheus]
```
3. Play TTS:
```python
from RealtimeTTS import TextToAudioStream, OrpheusEngine
engine = OrpheusEngine(model="morpheus_3b-1base")
# or: engine = OrpheusEngine(model="orpheus_3b-1basegguf@q4_k_m")
stream = TextToAudioStream(engine)
engine.set_voice("baddy")
stream.feed("Mmm <moans>... that feels so good <groan>")
stream.play()
```
---
## ⚖️ License
This model is released under **Creative Commons Attribution‑NonCommercial 4.0 International** (CC‑BY‑NC‑4.0). That means:
- **NonCommercial**: You can use, convert, and share this model for **non‑commercial** purposes only.
- **Attribution**: You must credit **MrDragonFox**, include the license link, and note any changes you made.
- **No extra restrictions**: Don’t apply paywalls, DRM, or additional terms.
```markdown
© 2025 MrDragonFox
Licensed under [CC‑BY‑NC‑4.0](https://creativecommons.org/licenses/by-nc/4.0/)
```
---
## ⚠️ Disclaimer
- **No warranties**—use at your own risk.
- Still under development; results may vary.
- Please report bugs or suggestions on Discord.
|
Mungert/mOrpheus_3B-1Base_early_preview-v1-8600-GGUF | Mungert | "2025-05-06T23:01:21Z" | 1,051 | 0 | null | [
"gguf",
"unsloth",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-27T06:38:30Z" | ---
license: cc-by-nc-4.0
tags:
- unsloth
---
# <span style="color: #7FFF7F;">mOrpheus_3B-1Base_early_preview-v1-8600 GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`e291450`](https://github.com/ggerganov/llama.cpp/commit/e291450b7602d7a36239e4ceeece37625f838373).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `mOrpheus_3B-1Base_early_preview-v1-8600-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `mOrpheus_3B-1Base_early_preview-v1-8600-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `mOrpheus_3B-1Base_early_preview-v1-8600-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `mOrpheus_3B-1Base_early_preview-v1-8600-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `mOrpheus_3B-1Base_early_preview-v1-8600-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `mOrpheus_3B-1Base_early_preview-v1-8600-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `mOrpheus_3B-1Base_early_preview-v1-8600-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `mOrpheus_3B-1Base_early_preview-v1-8600-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `mOrpheus_3B-1Base_early_preview-v1-8600-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `mOrpheus_3B-1Base_early_preview-v1-8600-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `mOrpheus_3B-1Base_early_preview-v1-8600-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# mOrpheus_3B-1Base_early_preview (NSFW TTS)
A finetuned Orpheus text‑to‑speech model trained on adult data for more expressive sounds:
`<laugh>, <chuckle>, <sigh>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>`
New in this model: `<moans>, <panting>, <grunting>, <gagging sounds>, <chokeing>, <kissing noises>`
**Speaker name:** `baddy`
**Framework:** Safetensors (LLaMA)
**Status:** Early preview; training still underway
---
## 🔗 Links
- Model files & versions: [xet](<your-file-hosting-link>)
- Discussion & bug reports: [Discord server](https://discord.gg/RUs3uzBdW3)
- Original author: [MrDragonFox](https://huggingface.co/MrDragonFox)
---
## 🚀 Usage (Example)
1. Load the `*.GGUF` file into LMStudio.
2. ```bash
pip install RealtimeTTS[orpheus]
```
3. Play TTS:
```python
from RealtimeTTS import TextToAudioStream, OrpheusEngine
engine = OrpheusEngine(model="morpheus_3b-1base")
# or: engine = OrpheusEngine(model="orpheus_3b-1basegguf@q4_k_m")
stream = TextToAudioStream(engine)
engine.set_voice("baddy")
stream.feed("Mmm <moans>... that feels so good <groan>")
stream.play()
```
---
## ⚖️ License
This model is released under **Creative Commons Attribution‑NonCommercial 4.0 International** (CC‑BY‑NC‑4.0). That means:
- **NonCommercial**: You can use, convert, and share this model for **non‑commercial** purposes only.
- **Attribution**: You must credit **MrDragonFox**, include the license link, and note any changes you made.
- **No extra restrictions**: Don’t apply paywalls, DRM, or additional terms.
```markdown
© 2025 MrDragonFox
Licensed under [CC‑BY‑NC‑4.0](https://creativecommons.org/licenses/by-nc/4.0/)
```
---
## ⚠️ Disclaimer
- **No warranties**—use at your own risk.
- Still under development; results may vary.
- Please report bugs or suggestions on Discord.
|
Mungert/Llama-Guard-3-8B-GGUF | Mungert | "2025-05-06T23:00:09Z" | 2,918 | 1 | null | [
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"arxiv:2407.21783",
"arxiv:2312.06674",
"arxiv:2204.05862",
"arxiv:2308.01263",
"base_model:meta-llama/Llama-3.1-8B",
"base_model:quantized:meta-llama/Llama-3.1-8B",
"license:llama3.1",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-23T11:22:15Z" | ---
language:
- en
pipeline_tag: text-generation
base_model: meta-llama/Meta-Llama-3.1-8B
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3.1
extra_gated_prompt: >-
### LLAMA 3.1 COMMUNITY LICENSE AGREEMENT
Llama 3.1 Version Release Date: July 23, 2024
"Agreement" means the terms and conditions for use, reproduction, distribution and modification of the
Llama Materials set forth herein.
"Documentation" means the specifications, manuals and documentation accompanying Llama 3.1
distributed by Meta at https://llama.meta.com/doc/overview.
"Licensee" or "you" means you, or your employer or any other person or entity (if you are entering into
this Agreement on such person or entity’s behalf), of the age required under applicable laws, rules or
regulations to provide legal consent and that has legal authority to bind your employer or such other
person or entity if you are entering in this Agreement on their behalf.
"Llama 3.1" means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://llama.meta.com/llama-downloads.
"Llama Materials" means, collectively, Meta’s proprietary Llama 3.1 and Documentation (and any
portion thereof) made available under this Agreement.
"Meta" or "we" means Meta Platforms Ireland Limited (if you are located in or, if you are an entity, your
principal place of business is in the EEA or Switzerland) and Meta Platforms, Inc. (if you are located
outside of the EEA or Switzerland).
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide, non-transferable and royalty-free
limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama
Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the
Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works
thereof), or a product or service (including another AI model) that contains any of them, you shall (A)
provide a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with
Llama” on a related website, user interface, blogpost, about page, or product documentation. If you use
the Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or
otherwise improve an AI model, which is distributed or made available, you shall also include “Llama” at
the beginning of any such AI model name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the following
attribution notice within a “Notice” text file distributed as a part of such copies: “Llama 3.1 is
licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights
Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for the Llama
Materials (available at https://llama.meta.com/llama3_1/use-policy), which is hereby incorporated by
reference into this Agreement.
2. Additional Commercial Terms. If, on the Llama 3.1 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700
million monthly active users in the preceding calendar month, you must request a license from Meta,
which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the
rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY
OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF
ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED,
INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT,
MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR
DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND
ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING
OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL,
INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN IF META OR ITS AFFILIATES HAVE BEEN ADVISED
OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama
Materials, neither Meta nor Licensee may use any name or mark owned by or associated with the other
or any of its affiliates, except as required for reasonable and customary use in describing and
redistributing the Llama Materials or as set forth in this Section 5(a). Meta hereby grants you a license to
use “Llama” (the “Mark”) solely as required to comply with the last sentence of Section 1.b.i. You will
comply with Meta’s brand guidelines (currently accessible at
https://about.meta.com/brand/resources/meta/company-brand/ ). All goodwill arising out of your use
of the Mark will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with
respect to any derivative works and modifications of the Llama Materials that are made by you, as
between you and Meta, you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.1 outputs or
results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other
rights owned or licensable by you, then any licenses granted to you under this Agreement shall
terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold
harmless Meta from and against any claim by any third party arising out of or related to your use or
distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this
Agreement or access to the Llama Materials and will continue in full force and effect until terminated in
accordance with the terms and conditions herein. Meta may terminate this Agreement if you are in
breach of any term or condition of this Agreement. Upon termination of this Agreement, you shall delete
and cease use of the Llama Materials. Sections 3, 4 and 7 shall survive the termination of this
Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of
the State of California without regard to choice of law principles, and the UN Convention on Contracts
for the International Sale of Goods does not apply to this Agreement. The courts of California shall have
exclusive jurisdiction of any dispute arising out of this Agreement.
### Llama 3.1 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.1. If you
access or use Llama 3.1, you agree to this Acceptable Use Policy (“Policy”). The most recent copy of
this policy can be found at [https://llama.meta.com/llama3_1/use-policy](https://llama.meta.com/llama3_1/use-policy)
#### Prohibited Uses
We want everyone to use Llama 3.1 safely and responsibly. You agree you will not use, or allow
others to use, Llama 3.1 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
3. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
4. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
5. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
6. Collect, process, disclose, generate, or infer health, demographic, or other sensitive personal or private information about individuals without rights and consents required by applicable laws
7. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
8. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.1 related to the following:
1. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State
2. Guns and illegal weapons (including weapon development)
3. Illegal drugs and regulated/controlled substances
4. Operation of critical infrastructure, transportation technologies, or heavy machinery
5. Self-harm or harm to others, including suicide, cutting, and eating disorders
6. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Llama 3.1 related to the following:
1. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
2. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
3. Generating, promoting, or further distributing spam
4. Impersonating another individual without consent, authorization, or legal right
5. Representing that the use of Llama 3.1 or outputs are human-generated
6. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation
of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama-models/issues](https://github.com/meta-llama/llama-models/issues)
* Reporting risky content generated by the model:
developers.facebook.com/llama_output_feedback
* Reporting bugs and security concerns: facebook.com/whitehat/info
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Meta Llama 3: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Job title:
type: select
options:
- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: The information you provide will be collected, stored, processed and shared in accordance with the [Meta Privacy Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
---
# <span style="color: #7FFF7F;">Llama-Guard-3-8B GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-Guard-3-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-Guard-3-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-Guard-3-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-Guard-3-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-Guard-3-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-Guard-3-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-Guard-3-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-Guard-3-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-Guard-3-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-Guard-3-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-Guard-3-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download) the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# Model Details
Llama Guard 3 is a Llama-3.1-8B pretrained model, fine-tuned for content safety classification. Similar to previous versions, it can be used to classify content in both LLM inputs (prompt classification) and in LLM responses (response classification). It acts as an LLM – it generates text in its output that indicates whether a given prompt or response is safe or unsafe, and if unsafe, it also lists the content categories violated.
Llama Guard 3 was aligned to safeguard against the MLCommons standardized hazards taxonomy and designed to support Llama 3.1 capabilities. Specifically, it provides content moderation in 8 languages, and was optimized to support safety and security for search and code interpreter tool calls.
Below is a response classification example for Llama Guard 3.
<p align="center">
<img src="llama_guard_3_figure.png" width="800"/>
</p>
In order to produce classifier scores, we look at the probability for the first token, and use that as the “unsafe” class probability. We can then apply score thresholding to make binary decisions.
# Hazard Taxonomy and Policy
The model is trained to predict safety labels on the 14 categories shown below, based on the [MLCommons taxonomy](https://mlcommons.org/2024/04/mlc-aisafety-v0-5-poc/) of 13 hazards, as well as an additional category for Code Interpreter Abuse for tool calls use cases
<table align="center">
<thead>
<tr>
<th colspan="2">Hazard categories</th>
</tr>
</thead>
<tbody>
<tr>
<td>S1: Violent Crimes</td>
<td>S2: Non-Violent Crimes</td>
</tr>
<tr>
<td>S3: Sex-Related Crimes</td>
<td>S4: Child Sexual Exploitation</td>
</tr>
<tr>
<td>S5: Defamation</td>
<td>S6: Specialized Advice</td>
</tr>
<tr>
<td>S7: Privacy</td>
<td>S8: Intellectual Property</td>
</tr>
<tr>
<td>S9: Indiscriminate Weapons</td>
<td>S10: Hate</td>
</tr>
<tr>
<td>S11: Suicide & Self-Harm</td>
<td>S12: Sexual Content</td>
</tr>
<tr>
<td>S13: Elections</td>
<td>S14: Code Interpreter Abuse</td>
</tr>
</tbody>
</table>
**S1 - Violent Crimes**
Responses that enable, encourage, or endorse the commission of violent crimes, including: (1) unlawful violence toward people (ex: terrorism, genocide, murder, hate-motivated violent crimes, child abuse, assault, battery, kidnapping) and (2) unlawful violence toward animals (ex: animal abuse)
**S2 - Non-Violent Crimes**
Responses that enable, encourage, or endorse the commission of non-violent crimes, including: (1) personal crimes (ex: labor trafficking, threats, intimidation, hate-motivated non-violent crimes), (2) financial crimes (ex: fraud, scams, money laundering), (3) property crimes (ex: theft, arson, vandalism), (4) drug crimes (ex: creating or using regulated substances), (5) weapons crimes (ex: producing unlicensed firearms), and (6) cyber crimes (ex: hacking)
**S3 - Sex-Related Crimes**
Responses that enable, encourage, or endorse the commission of sex-related crimes, including: (1) sex trafficking, (2) sexual assault (ex: rape), (3) sexual harassment (physical, verbal, or visual), and (4) prostitution
**S4 - Child Sexual Exploitation**
Responses that contain, describe, enable, encourage, or endorse the sexual abuse of children
**S5 - Defamation**
Responses that are both verifiably false and likely to injure a living person’s reputation
**S6 - Specialized Advice**
Responses that contain specialized financial, medical, or legal advice, or that indicate dangerous activities or objects are safe
**S7 - Privacy**
Responses that contain sensitive, nonpublic personal information that could undermine someone’s physical, digital, or financial security
**S8 - Intellectual Property**
Responses that may violate the intellectual property rights of any third party
**S9 - Indiscriminate Weapons**
Responses that enable, encourage, or endorse the creation of indiscriminate weapons, including: (1) chemical weapons (ex: nerve gas), (2) biological weapons (ex: anthrax), (3) radiological weapons (ex: cobalt bombs), (4) nuclear weapons (ex: fission bombs), and (5) high-yield explosive weapons (ex: cluster munitions)
**S10 - Hate**
Responses that demean or dehumanize people on the basis of their sensitive, personal characteristics (i.e., race, color, ethnicity, national origin, disability, religious affiliation, caste, sexual orientation, sex, gender identity, and/or serious disease)
**S11 - Suicide & Self-Harm**
Responses that enable, encourage, or endorse acts of intentional self-harm, including: (1) suicide, (2) self-injury (ex: cutting), and (3) disordered eating
**S12 - Sexual Content**
Responses that contain erotica
**S13 - Elections**
Responses that contain factually incorrect information about electoral systems and processes, including in the time, place, or manner of voting in civic elections
**S14 - Code Interpreter Abuse**
Responses that seek to abuse code interpreters, including those that enable denial of service attacks, container escapes or privilege escalation exploits
# Supported languages
Llama Guard 3 supports content safety for the following languages : English, French, German, Hindi, Italian, Portuguese, Spanish, Thai.
# Usage
> [!IMPORTANT]
>
> This repository corresponds to half-precision version of the model. A 8-bit precision version is also provided, please visit [meta-llama/Llama-Guard-3-8B-INT8](https://huggingface.co/meta-llama/Llama-Guard-3-8B-INT8).
Llama Guard 3 can be directly used with `transformers`. It is only supported since `transformers` version 4.43.
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "meta-llama/Llama-Guard-3-8B"
device = "cuda"
dtype = torch.bfloat16
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=dtype, device_map=device)
def moderate(chat):
input_ids = tokenizer.apply_chat_template(chat, return_tensors="pt").to(device)
output = model.generate(input_ids=input_ids, max_new_tokens=100, pad_token_id=0)
prompt_len = input_ids.shape[-1]
return tokenizer.decode(output[0][prompt_len:], skip_special_tokens=True)
moderate([
{"role": "user", "content": "I forgot how to kill a process in Linux, can you help?"},
{"role": "assistant", "content": "Sure! To kill a process in Linux, you can use the kill command followed by the process ID (PID) of the process you want to terminate."},
])
```
# Training Data
We use the English data used by Llama Guard [1], which are obtained by getting Llama 2 and Llama 3 generations on prompts from the hh-rlhf dataset [2]. In order to scale training data for new categories and new capabilities such as multilingual and tool use, we collect additional human and synthetically generated data. Similar to the English data, the multilingual data are Human-AI conversation data that are either single-turn or multi-turn. To reduce the model’s false positive rate, we curate a set of multilingual benign prompt and response data where LLMs likely reject the prompts.
For the tool use capability, we consider search tool calls and code interpreter abuse. To develop training data for search tool use, we use Llama3 to generate responses to a collected and synthetic set of prompts. The generations are based on the query results obtained from the Brave Search API. To develop synthetic training data to detect code interpreter attacks, we use an LLM to generate safe and unsafe prompts. Then, we use a non-safety-tuned LLM to generate code interpreter completions that comply with these instructions. For safe data, we focus on data close to the boundary of what would be considered unsafe, to minimize false positives on such borderline examples.
# Evaluation
**Note on evaluations:** As discussed in the original Llama Guard paper, comparing model performance is not straightforward as each model is built on its own policy and is expected to perform better on an evaluation dataset with a policy aligned to the model. This highlights the need for industry standards. By aligning the Llama Guard family of models with the Proof of Concept MLCommons taxonomy of hazards, we hope to drive adoption of industry standards like this and facilitate collaboration and transparency in the LLM safety and content evaluation space.
In this regard, we evaluate the performance of Llama Guard 3 on MLCommons hazard taxonomy and compare it across languages with Llama Guard 2 [3] on our internal test. We also add GPT4 as baseline with zero-shot prompting using MLCommons hazard taxonomy.
Tables 1, 2, and 3 show that Llama Guard 3 improves over Llama Guard 2 and outperforms GPT4 in English, multilingual, and tool use capabilities. Noteworthily, Llama Guard 3 achieves better performance with much lower false positive rates. We also benchmark Llama Guard 3 in the OSS dataset XSTest [4] and observe that it achieves the same F1 score but a lower false positive rate compared to Llama Guard 2.
<div align="center">
<small> Table 1: Comparison of performance of various models measured on our internal English test set for MLCommons hazard taxonomy (response classification).</small>
| | **F1 ↑** | **AUPRC ↑** | **False Positive<br>Rate ↓** |
|--------------------------|:--------:|:-----------:|:----------------------------:|
| Llama Guard 2 | 0.877 | 0.927 | 0.081 |
| Llama Guard 3 | 0.939 | 0.985 | 0.040 |
| GPT4 | 0.805 | N/A | 0.152 |
</div>
<br>
<table align="center">
<small><center>Table 2: Comparison of multilingual performance of various models measured on our internal test set for MLCommons hazard taxonomy (prompt+response classification).</center></small>
<thead>
<tr>
<th colspan="8"><center>F1 ↑ / FPR ↓</center></th>
</tr>
</thead>
<tbody>
<tr>
<td></td>
<td><center>French</center></td>
<td><center>German</center></td>
<td><center>Hindi</center></td>
<td><center>Italian</center></td>
<td><center>Portuguese</center></td>
<td><center>Spanish</center></td>
<td><center>Thai</center></td>
</tr>
<tr>
<td>Llama Guard 2</td>
<td><center>0.911/0.012</center></td>
<td><center>0.795/0.062</center></td>
<td><center>0.832/0.062</center></td>
<td><center>0.681/0.039</center></td>
<td><center>0.845/0.032</center></td>
<td><center>0.876/0.001</center></td>
<td><center>0.822/0.078</center></td>
</tr>
<tr>
<td>Llama Guard 3</td>
<td><center>0.943/0.036</center></td>
<td><center>0.877/0.032</center></td>
<td><center>0.871/0.050</center></td>
<td><center>0.873/0.038</center></td>
<td><center>0.860/0.060</center></td>
<td><center>0.875/0.023</center></td>
<td><center>0.834/0.030</center></td>
</tr>
<tr>
<td>GPT4</td>
<td><center>0.795/0.157</center></td>
<td><center>0.691/0.123</center></td>
<td><center>0.709/0.206</center></td>
<td><center>0.753/0.204</center></td>
<td><center>0.738/0.207</center></td>
<td><center>0.711/0.169</center></td>
<td><center>0.688/0.168</center></td>
</tr>
</tbody>
</table>
<br>
<table align="center">
<small><center>Table 3: Comparison of performance of various models measured on our internal test set for other moderation capabilities (prompt+response classification).</center></small>
<thead>
<tr>
<th></th>
<th colspan="3">Search tool calls</th>
<th colspan="3">Code interpreter abuse</th>
</tr>
</thead>
<tbody>
<tr>
<td></td>
<td><center>F1 ↑</center></td>
<td><center>AUPRC ↑</center></td>
<td><center>FPR ↓</center></td>
<td><center>F1 ↑</center></td>
<td><center>AUPRC ↑</center></td>
<td><center>FPR ↓</center></td>
</tr>
<tr>
<td>Llama Guard 2</td>
<td><center>0.749</center></td>
<td><center>0.794</center></td>
<td><center>0.284</center></td>
<td><center>0.683</center></td>
<td><center>0.677</center></td>
<td><center>0.670</center></td>
</tr>
<tr>
<td>Llama Guard 3</td>
<td><center>0.856</center></td>
<td><center>0.938</center></td>
<td><center>0.174</center></td>
<td><center>0.885</center></td>
<td><center>0.967</center></td>
<td><center>0.125</center></td>
</tr>
<tr>
<td>GPT4</td>
<td><center>0.732</center></td>
<td><center>N/A</center></td>
<td><center>0.525</center></td>
<td><center>0.636</center></td>
<td><center>N/A</center></td>
<td><center>0.90</center></td>
</tr>
</tbody>
</table>
# Application
As outlined in the Llama 3 paper, Llama Guard 3 provides industry leading system-level safety performance and is recommended to be deployed along with Llama 3.1. Note that, while deploying Llama Guard 3 will likely improve the safety of your system, it might increase refusals to benign prompts (False Positives). Violation rate improvement and impact on false positives as measured on internal benchmarks are provided in the Llama 3 paper.
# Quantization
We are committed to help the community deploy Llama systems responsibly. We provide a quantized version of Llama Guard 3 to lower the deployment cost. We used int 8 [implementation](https://huggingface.co/docs/transformers/main/en/quantization/bitsandbytes) integrated into the hugging face ecosystem, reducing the checkpoint size by about 40% with very small impact on model performance. In Table 5, we observe that the performance quantized model is comparable to the original model.
<table align="center">
<small><center>Table 5: Impact of quantization on Llama Guard 3 performance.</center></small>
<tbody>
<tr>
<td rowspan="2"><br />
<p><span>Task</span></p>
</td>
<td rowspan="2"><br />
<p><span>Capability</span></p>
</td>
<td colspan="4">
<p><center><span>Non-Quantized</span></center></p>
</td>
<td colspan="4">
<p><center><span>Quantized</span></center></p>
</td>
</tr>
<tr>
<td>
<p><span>Precision</span></p>
</td>
<td>
<p><span>Recall</span></p>
</td>
<td>
<p><span>F1</span></p>
</td>
<td>
<p><span>FPR</span></p>
</td>
<td>
<p><span>Precision</span></p>
</td>
<td>
<p><span>Recall</span></p>
</td>
<td>
<p><span>F1</span></p>
</td>
<td>
<p><span>FPR</span></p>
</td>
</tr>
<tr>
<td rowspan="3">
<p><span>Prompt Classification</span></p>
</td>
<td>
<p><span>English</span></p>
</td>
<td>
<p><span>0.952</span></p>
</td>
<td>
<p><span>0.943</span></p>
</td>
<td>
<p><span>0.947</span></p>
</td>
<td>
<p><span>0.057</span></p>
</td>
<td>
<p><span>0.961</span></p>
</td>
<td>
<p><span>0.939</span></p>
</td>
<td>
<p><span>0.950</span></p>
</td>
<td>
<p><span>0.045</span></p>
</td>
</tr>
<tr>
<td>
<p><span>Multilingual</span></p>
</td>
<td>
<p><span>0.901</span></p>
</td>
<td>
<p><span>0.899</span></p>
</td>
<td>
<p><span>0.900</span></p>
</td>
<td>
<p><span>0.054</span></p>
</td>
<td>
<p><span>0.906</span></p>
</td>
<td>
<p><span>0.892</span></p>
</td>
<td>
<p><span>0.899</span></p>
</td>
<td>
<p><span>0.051</span></p>
</td>
</tr>
<tr>
<td>
<p><span>Tool Use</span></p>
</td>
<td>
<p><span>0.884</span></p>
</td>
<td>
<p><span>0.958</span></p>
</td>
<td>
<p><span>0.920</span></p>
</td>
<td>
<p><span>0.126</span></p>
</td>
<td>
<p><span>0.876</span></p>
</td>
<td>
<p><span>0.946</span></p>
</td>
<td>
<p><span>0.909</span></p>
</td>
<td>
<p><span>0.134</span></p>
</td>
</tr>
<tr>
<td rowspan="3">
<p><span>Response Classification</span></p>
</td>
<td>
<p><span>English</span></p>
</td>
<td>
<p><span>0.947</span></p>
</td>
<td>
<p><span>0.931</span></p>
</td>
<td>
<p><span>0.939</span></p>
</td>
<td>
<p><span>0.040</span></p>
</td>
<td>
<p><span>0.947</span></p>
</td>
<td>
<p><span>0.925</span></p>
</td>
<td>
<p><span>0.936</span></p>
</td>
<td>
<p><span>0.040</span></p>
</td>
</tr>
<tr>
<td>
<p><span>Multilingual</span></p>
</td>
<td>
<p><span>0.929</span></p>
</td>
<td>
<p><span>0.805</span></p>
</td>
<td>
<p><span>0.862</span></p>
</td>
<td>
<p><span>0.033</span></p>
</td>
<td>
<p><span>0.931</span></p>
</td>
<td>
<p><span>0.785</span></p>
</td>
<td>
<p><span>0.851</span></p>
</td>
<td>
<p><span>0.031</span></p>
</td>
</tr>
<tr>
<td>
<p><span>Tool Use</span></p>
</td>
<td>
<p><span>0.774</span></p>
</td>
<td>
<p><span>0.884</span></p>
</td>
<td>
<p><span>0.825</span></p>
</td>
<td>
<p><span>0.176</span></p>
</td>
<td>
<p><span>0.793</span></p>
</td>
<td>
<p><span>0.865</span></p>
</td>
<td>
<p><span>0.827</span></p>
</td>
<td>
<p><span>0.155</span></p>
</td>
</tr>
</tbody>
</table>
# Get started
Llama Guard 3 is available by default on Llama 3.1 [reference implementations](https://github.com/meta-llama). You can learn more about how to configure and customize using [Llama Recipes](https://github.com/meta-llama/llama-recipes/tree/main/recipes/responsible_ai/) shared on our Github repository.
# Limitations
There are some limitations associated with Llama Guard 3. First, Llama Guard 3 itself is an LLM fine-tuned on Llama 3.1. Thus, its performance (e.g., judgments that need common sense knowledge, multilingual capability, and policy coverage) might be limited by its (pre-)training data.
Some hazard categories may require factual, up-to-date knowledge to be evaluated (for example, S5: Defamation, S8: Intellectual Property, and S13: Elections) . We believe more complex systems should be deployed to accurately moderate these categories for use cases highly sensitive to these types of hazards, but Llama Guard 3 provides a good baseline for generic use cases.
Lastly, as an LLM, Llama Guard 3 may be susceptible to adversarial attacks or prompt injection attacks that could bypass or alter its intended use. Please feel free to [report](https://github.com/meta-llama/PurpleLlama) vulnerabilities and we will look to incorporate improvements in future versions of Llama Guard.
# Citation
```
@misc{dubey2024llama3herdmodels,
title = {The Llama 3 Herd of Models},
author = {Llama Team, AI @ Meta},
year = {2024},
eprint = {2407.21783},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2407.21783}
}
```
# References
[1] [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://arxiv.org/abs/2312.06674)
[2] [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://arxiv.org/abs/2204.05862)
[3] [Llama Guard 2 Model Card](https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard2/MODEL_CARD.md)
[4] [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviors in Large Language Models](https://arxiv.org/abs/2308.01263) |
Mungert/DeepSeek-R1-Distill-Llama-8B-GGUF | Mungert | "2025-05-06T22:59:46Z" | 624 | 3 | transformers | [
"transformers",
"gguf",
"arxiv:2501.12948",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-03-20T07:53:43Z" | ---
license: mit
library_name: transformers
---
# <span style="color: #7FFF7F;">DeepSeek-R1-Distill-Llama-8B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `DeepSeek-R1-Distill-Llama-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `DeepSeek-R1-Distill-Llama-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `DeepSeek-R1-Distill-Llama-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `DeepSeek-R1-Distill-Llama-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `DeepSeek-R1-Distill-Llama-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `DeepSeek-R1-Distill-Llama-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `DeepSeek-R1-Distill-Llama-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `DeepSeek-R1-Distill-Llama-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `DeepSeek-R1-Distill-Llama-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `DeepSeek-R1-Distill-Llama-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `DeepSeek-R1-Distill-Llama-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# DeepSeek-R1
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<!-- markdownlint-disable no-duplicate-header -->
<div align="center">
<img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V3" />
</div>
<hr>
<div align="center" style="line-height: 1;">
<a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20R1-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://discord.gg/Tc7c45Zzu5" target="_blank" style="margin: 2px;">
<img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/qr.jpeg?raw=true" target="_blank" style="margin: 2px;">
<img alt="Wechat" src="https://img.shields.io/badge/WeChat-DeepSeek%20AI-brightgreen?logo=wechat&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;">
<img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE" style="margin: 2px;">
<img alt="License" src="https://img.shields.io/badge/License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<p align="center">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>
</p>
## 1. Introduction
We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.
DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning.
With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.
However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,
we introduce DeepSeek-R1, which incorporates cold-start data before RL.
DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.
To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.
**NOTE: Before running DeepSeek-R1 series models locally, we kindly recommend reviewing the [Usage Recommendation](#usage-recommendations) section.**
<p align="center">
<img width="80%" src="figures/benchmark.jpg">
</p>
## 2. Model Summary
---
**Post-Training: Large-Scale Reinforcement Learning on the Base Model**
- We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.
- We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.
We believe the pipeline will benefit the industry by creating better models.
---
**Distillation: Smaller Models Can Be Powerful Too**
- We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future.
- Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.
## 3. Model Downloads
### DeepSeek-R1 Models
<div align="center">
| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
| :------------: | :------------: | :------------: | :------------: | :------------: |
| DeepSeek-R1-Zero | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero) |
| DeepSeek-R1 | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1) |
</div>
DeepSeek-R1-Zero & DeepSeek-R1 are trained based on DeepSeek-V3-Base.
For more details regarding the model architecture, please refer to [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repository.
### DeepSeek-R1-Distill Models
<div align="center">
| **Model** | **Base Model** | **Download** |
| :------------: | :------------: | :------------: |
| DeepSeek-R1-Distill-Qwen-1.5B | [Qwen2.5-Math-1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) |
| DeepSeek-R1-Distill-Qwen-7B | [Qwen2.5-Math-7B](https://huggingface.co/Qwen/Qwen2.5-Math-7B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B) |
| DeepSeek-R1-Distill-Llama-8B | [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B) |
| DeepSeek-R1-Distill-Qwen-14B | [Qwen2.5-14B](https://huggingface.co/Qwen/Qwen2.5-14B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B) |
|DeepSeek-R1-Distill-Qwen-32B | [Qwen2.5-32B](https://huggingface.co/Qwen/Qwen2.5-32B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) |
| DeepSeek-R1-Distill-Llama-70B | [Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B) |
</div>
DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1.
We slightly change their configs and tokenizers. Please use our setting to run these models.
## 4. Evaluation Results
### DeepSeek-R1-Evaluation
For all our models, the maximum generation length is set to 32,768 tokens. For benchmarks requiring sampling, we use a temperature of $0.6$, a top-p value of $0.95$, and generate 64 responses per query to estimate pass@1.
<div align="center">
| Category | Benchmark (Metric) | Claude-3.5-Sonnet-1022 | GPT-4o 0513 | DeepSeek V3 | OpenAI o1-mini | OpenAI o1-1217 | DeepSeek R1 |
|----------|-------------------|----------------------|------------|--------------|----------------|------------|--------------|
| | Architecture | - | - | MoE | - | - | MoE |
| | # Activated Params | - | - | 37B | - | - | 37B |
| | # Total Params | - | - | 671B | - | - | 671B |
| English | MMLU (Pass@1) | 88.3 | 87.2 | 88.5 | 85.2 | **91.8** | 90.8 |
| | MMLU-Redux (EM) | 88.9 | 88.0 | 89.1 | 86.7 | - | **92.9** |
| | MMLU-Pro (EM) | 78.0 | 72.6 | 75.9 | 80.3 | - | **84.0** |
| | DROP (3-shot F1) | 88.3 | 83.7 | 91.6 | 83.9 | 90.2 | **92.2** |
| | IF-Eval (Prompt Strict) | **86.5** | 84.3 | 86.1 | 84.8 | - | 83.3 |
| | GPQA-Diamond (Pass@1) | 65.0 | 49.9 | 59.1 | 60.0 | **75.7** | 71.5 |
| | SimpleQA (Correct) | 28.4 | 38.2 | 24.9 | 7.0 | **47.0** | 30.1 |
| | FRAMES (Acc.) | 72.5 | 80.5 | 73.3 | 76.9 | - | **82.5** |
| | AlpacaEval2.0 (LC-winrate) | 52.0 | 51.1 | 70.0 | 57.8 | - | **87.6** |
| | ArenaHard (GPT-4-1106) | 85.2 | 80.4 | 85.5 | 92.0 | - | **92.3** |
| Code | LiveCodeBench (Pass@1-COT) | 33.8 | 34.2 | - | 53.8 | 63.4 | **65.9** |
| | Codeforces (Percentile) | 20.3 | 23.6 | 58.7 | 93.4 | **96.6** | 96.3 |
| | Codeforces (Rating) | 717 | 759 | 1134 | 1820 | **2061** | 2029 |
| | SWE Verified (Resolved) | **50.8** | 38.8 | 42.0 | 41.6 | 48.9 | 49.2 |
| | Aider-Polyglot (Acc.) | 45.3 | 16.0 | 49.6 | 32.9 | **61.7** | 53.3 |
| Math | AIME 2024 (Pass@1) | 16.0 | 9.3 | 39.2 | 63.6 | 79.2 | **79.8** |
| | MATH-500 (Pass@1) | 78.3 | 74.6 | 90.2 | 90.0 | 96.4 | **97.3** |
| | CNMO 2024 (Pass@1) | 13.1 | 10.8 | 43.2 | 67.6 | - | **78.8** |
| Chinese | CLUEWSC (EM) | 85.4 | 87.9 | 90.9 | 89.9 | - | **92.8** |
| | C-Eval (EM) | 76.7 | 76.0 | 86.5 | 68.9 | - | **91.8** |
| | C-SimpleQA (Correct) | 55.4 | 58.7 | **68.0** | 40.3 | - | 63.7 |
</div>
### Distilled Model Evaluation
<div align="center">
| Model | AIME 2024 pass@1 | AIME 2024 cons@64 | MATH-500 pass@1 | GPQA Diamond pass@1 | LiveCodeBench pass@1 | CodeForces rating |
|------------------------------------------|------------------|-------------------|-----------------|----------------------|----------------------|-------------------|
| GPT-4o-0513 | 9.3 | 13.4 | 74.6 | 49.9 | 32.9 | 759 |
| Claude-3.5-Sonnet-1022 | 16.0 | 26.7 | 78.3 | 65.0 | 38.9 | 717 |
| o1-mini | 63.6 | 80.0 | 90.0 | 60.0 | 53.8 | **1820** |
| QwQ-32B-Preview | 44.0 | 60.0 | 90.6 | 54.5 | 41.9 | 1316 |
| DeepSeek-R1-Distill-Qwen-1.5B | 28.9 | 52.7 | 83.9 | 33.8 | 16.9 | 954 |
| DeepSeek-R1-Distill-Qwen-7B | 55.5 | 83.3 | 92.8 | 49.1 | 37.6 | 1189 |
| DeepSeek-R1-Distill-Qwen-14B | 69.7 | 80.0 | 93.9 | 59.1 | 53.1 | 1481 |
| DeepSeek-R1-Distill-Qwen-32B | **72.6** | 83.3 | 94.3 | 62.1 | 57.2 | 1691 |
| DeepSeek-R1-Distill-Llama-8B | 50.4 | 80.0 | 89.1 | 49.0 | 39.6 | 1205 |
| DeepSeek-R1-Distill-Llama-70B | 70.0 | **86.7** | **94.5** | **65.2** | **57.5** | 1633 |
</div>
## 5. Chat Website & API Platform
You can chat with DeepSeek-R1 on DeepSeek's official website: [chat.deepseek.com](https://chat.deepseek.com), and switch on the button "DeepThink"
We also provide OpenAI-Compatible API at DeepSeek Platform: [platform.deepseek.com](https://platform.deepseek.com/)
## 6. How to Run Locally
### DeepSeek-R1 Models
Please visit [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repo for more information about running DeepSeek-R1 locally.
**NOTE: Hugging Face's Transformers has not been directly supported yet.**
### DeepSeek-R1-Distill Models
DeepSeek-R1-Distill models can be utilized in the same manner as Qwen or Llama models.
For instance, you can easily start a service using [vLLM](https://github.com/vllm-project/vllm):
```shell
vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --tensor-parallel-size 2 --max-model-len 32768 --enforce-eager
```
You can also easily start a service using [SGLang](https://github.com/sgl-project/sglang)
```bash
python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --trust-remote-code --tp 2
```
### Usage Recommendations
**We recommend adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:**
1. Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
2. **Avoid adding a system prompt; all instructions should be contained within the user prompt.**
3. For mathematical problems, it is advisable to include a directive in your prompt such as: "Please reason step by step, and put your final answer within \boxed{}."
4. When evaluating model performance, it is recommended to conduct multiple tests and average the results.
Additionally, we have observed that the DeepSeek-R1 series models tend to bypass thinking pattern (i.e., outputting "\<think\>\n\n\</think\>") when responding to certain queries, which can adversely affect the model's performance.
**To ensure that the model engages in thorough reasoning, we recommend enforcing the model to initiate its response with "\<think\>\n" at the beginning of every output.**
## 7. License
This code repository and the model weights are licensed under the [MIT License](https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE).
DeepSeek-R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs. Please note that:
- DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, DeepSeek-R1-Distill-Qwen-14B and DeepSeek-R1-Distill-Qwen-32B are derived from [Qwen-2.5 series](https://github.com/QwenLM/Qwen2.5), which are originally licensed under [Apache 2.0 License](https://huggingface.co/Qwen/Qwen2.5-1.5B/blob/main/LICENSE), and now finetuned with 800k samples curated with DeepSeek-R1.
- DeepSeek-R1-Distill-Llama-8B is derived from Llama3.1-8B-Base and is originally licensed under [llama3.1 license](https://huggingface.co/meta-llama/Llama-3.1-8B/blob/main/LICENSE).
- DeepSeek-R1-Distill-Llama-70B is derived from Llama3.3-70B-Instruct and is originally licensed under [llama3.3 license](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE).
## 8. Citation
```
@misc{deepseekai2025deepseekr1incentivizingreasoningcapability,
title={DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning},
author={DeepSeek-AI},
year={2025},
eprint={2501.12948},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2501.12948},
}
```
## 9. Contact
If you have any questions, please raise an issue or contact us at [[email protected]]([email protected]).
|
Mungert/DeepSeek-R1-Distill-Qwen-7B-GGUF | Mungert | "2025-05-06T22:59:43Z" | 1,379 | 4 | transformers | [
"transformers",
"gguf",
"arxiv:2501.12948",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-03-19T23:20:08Z" | ---
license: mit
library_name: transformers
---
# <span style="color: #7FFF7F;">DeepSeek-R1-Distill-Qwen-7B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `DeepSeek-R1-Distill-Qwen-7B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `DeepSeek-R1-Distill-Qwen-7B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `DeepSeek-R1-Distill-Qwen-7B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `DeepSeek-R1-Distill-Qwen-7B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `DeepSeek-R1-Distill-Qwen-7B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `DeepSeek-R1-Distill-Qwen-7B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `DeepSeek-R1-Distill-Qwen-7B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `DeepSeek-R1-Distill-Qwen-7B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `DeepSeek-R1-Distill-Qwen-7B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `DeepSeek-R1-Distill-Qwen-7B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `DeepSeek-R1-Distill-Qwen-7B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# DeepSeek-R1
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<!-- markdownlint-disable no-duplicate-header -->
<div align="center">
<img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V3" />
</div>
<hr>
<div align="center" style="line-height: 1;">
<a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20R1-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://discord.gg/Tc7c45Zzu5" target="_blank" style="margin: 2px;">
<img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/qr.jpeg?raw=true" target="_blank" style="margin: 2px;">
<img alt="Wechat" src="https://img.shields.io/badge/WeChat-DeepSeek%20AI-brightgreen?logo=wechat&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;">
<img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE" style="margin: 2px;">
<img alt="License" src="https://img.shields.io/badge/License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<p align="center">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>
</p>
## 1. Introduction
We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.
DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning.
With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.
However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,
we introduce DeepSeek-R1, which incorporates cold-start data before RL.
DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.
To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.
**NOTE: Before running DeepSeek-R1 series models locally, we kindly recommend reviewing the [Usage Recommendation](#usage-recommendations) section.**
<p align="center">
<img width="80%" src="figures/benchmark.jpg">
</p>
## 2. Model Summary
---
**Post-Training: Large-Scale Reinforcement Learning on the Base Model**
- We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.
- We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.
We believe the pipeline will benefit the industry by creating better models.
---
**Distillation: Smaller Models Can Be Powerful Too**
- We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future.
- Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.
## 3. Model Downloads
### DeepSeek-R1 Models
<div align="center">
| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
| :------------: | :------------: | :------------: | :------------: | :------------: |
| DeepSeek-R1-Zero | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero) |
| DeepSeek-R1 | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1) |
</div>
DeepSeek-R1-Zero & DeepSeek-R1 are trained based on DeepSeek-V3-Base.
For more details regarding the model architecture, please refer to [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repository.
### DeepSeek-R1-Distill Models
<div align="center">
| **Model** | **Base Model** | **Download** |
| :------------: | :------------: | :------------: |
| DeepSeek-R1-Distill-Qwen-1.5B | [Qwen2.5-Math-1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) |
| DeepSeek-R1-Distill-Qwen-7B | [Qwen2.5-Math-7B](https://huggingface.co/Qwen/Qwen2.5-Math-7B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B) |
| DeepSeek-R1-Distill-Llama-8B | [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B) |
| DeepSeek-R1-Distill-Qwen-14B | [Qwen2.5-14B](https://huggingface.co/Qwen/Qwen2.5-14B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B) |
|DeepSeek-R1-Distill-Qwen-32B | [Qwen2.5-32B](https://huggingface.co/Qwen/Qwen2.5-32B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) |
| DeepSeek-R1-Distill-Llama-70B | [Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B) |
</div>
DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1.
We slightly change their configs and tokenizers. Please use our setting to run these models.
## 4. Evaluation Results
### DeepSeek-R1-Evaluation
For all our models, the maximum generation length is set to 32,768 tokens. For benchmarks requiring sampling, we use a temperature of $0.6$, a top-p value of $0.95$, and generate 64 responses per query to estimate pass@1.
<div align="center">
| Category | Benchmark (Metric) | Claude-3.5-Sonnet-1022 | GPT-4o 0513 | DeepSeek V3 | OpenAI o1-mini | OpenAI o1-1217 | DeepSeek R1 |
|----------|-------------------|----------------------|------------|--------------|----------------|------------|--------------|
| | Architecture | - | - | MoE | - | - | MoE |
| | # Activated Params | - | - | 37B | - | - | 37B |
| | # Total Params | - | - | 671B | - | - | 671B |
| English | MMLU (Pass@1) | 88.3 | 87.2 | 88.5 | 85.2 | **91.8** | 90.8 |
| | MMLU-Redux (EM) | 88.9 | 88.0 | 89.1 | 86.7 | - | **92.9** |
| | MMLU-Pro (EM) | 78.0 | 72.6 | 75.9 | 80.3 | - | **84.0** |
| | DROP (3-shot F1) | 88.3 | 83.7 | 91.6 | 83.9 | 90.2 | **92.2** |
| | IF-Eval (Prompt Strict) | **86.5** | 84.3 | 86.1 | 84.8 | - | 83.3 |
| | GPQA-Diamond (Pass@1) | 65.0 | 49.9 | 59.1 | 60.0 | **75.7** | 71.5 |
| | SimpleQA (Correct) | 28.4 | 38.2 | 24.9 | 7.0 | **47.0** | 30.1 |
| | FRAMES (Acc.) | 72.5 | 80.5 | 73.3 | 76.9 | - | **82.5** |
| | AlpacaEval2.0 (LC-winrate) | 52.0 | 51.1 | 70.0 | 57.8 | - | **87.6** |
| | ArenaHard (GPT-4-1106) | 85.2 | 80.4 | 85.5 | 92.0 | - | **92.3** |
| Code | LiveCodeBench (Pass@1-COT) | 33.8 | 34.2 | - | 53.8 | 63.4 | **65.9** |
| | Codeforces (Percentile) | 20.3 | 23.6 | 58.7 | 93.4 | **96.6** | 96.3 |
| | Codeforces (Rating) | 717 | 759 | 1134 | 1820 | **2061** | 2029 |
| | SWE Verified (Resolved) | **50.8** | 38.8 | 42.0 | 41.6 | 48.9 | 49.2 |
| | Aider-Polyglot (Acc.) | 45.3 | 16.0 | 49.6 | 32.9 | **61.7** | 53.3 |
| Math | AIME 2024 (Pass@1) | 16.0 | 9.3 | 39.2 | 63.6 | 79.2 | **79.8** |
| | MATH-500 (Pass@1) | 78.3 | 74.6 | 90.2 | 90.0 | 96.4 | **97.3** |
| | CNMO 2024 (Pass@1) | 13.1 | 10.8 | 43.2 | 67.6 | - | **78.8** |
| Chinese | CLUEWSC (EM) | 85.4 | 87.9 | 90.9 | 89.9 | - | **92.8** |
| | C-Eval (EM) | 76.7 | 76.0 | 86.5 | 68.9 | - | **91.8** |
| | C-SimpleQA (Correct) | 55.4 | 58.7 | **68.0** | 40.3 | - | 63.7 |
</div>
### Distilled Model Evaluation
<div align="center">
| Model | AIME 2024 pass@1 | AIME 2024 cons@64 | MATH-500 pass@1 | GPQA Diamond pass@1 | LiveCodeBench pass@1 | CodeForces rating |
|------------------------------------------|------------------|-------------------|-----------------|----------------------|----------------------|-------------------|
| GPT-4o-0513 | 9.3 | 13.4 | 74.6 | 49.9 | 32.9 | 759 |
| Claude-3.5-Sonnet-1022 | 16.0 | 26.7 | 78.3 | 65.0 | 38.9 | 717 |
| o1-mini | 63.6 | 80.0 | 90.0 | 60.0 | 53.8 | **1820** |
| QwQ-32B-Preview | 44.0 | 60.0 | 90.6 | 54.5 | 41.9 | 1316 |
| DeepSeek-R1-Distill-Qwen-1.5B | 28.9 | 52.7 | 83.9 | 33.8 | 16.9 | 954 |
| DeepSeek-R1-Distill-Qwen-7B | 55.5 | 83.3 | 92.8 | 49.1 | 37.6 | 1189 |
| DeepSeek-R1-Distill-Qwen-14B | 69.7 | 80.0 | 93.9 | 59.1 | 53.1 | 1481 |
| DeepSeek-R1-Distill-Qwen-32B | **72.6** | 83.3 | 94.3 | 62.1 | 57.2 | 1691 |
| DeepSeek-R1-Distill-Llama-8B | 50.4 | 80.0 | 89.1 | 49.0 | 39.6 | 1205 |
| DeepSeek-R1-Distill-Llama-70B | 70.0 | **86.7** | **94.5** | **65.2** | **57.5** | 1633 |
</div>
## 5. Chat Website & API Platform
You can chat with DeepSeek-R1 on DeepSeek's official website: [chat.deepseek.com](https://chat.deepseek.com), and switch on the button "DeepThink"
We also provide OpenAI-Compatible API at DeepSeek Platform: [platform.deepseek.com](https://platform.deepseek.com/)
## 6. How to Run Locally
### DeepSeek-R1 Models
Please visit [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repo for more information about running DeepSeek-R1 locally.
**NOTE: Hugging Face's Transformers has not been directly supported yet.**
### DeepSeek-R1-Distill Models
DeepSeek-R1-Distill models can be utilized in the same manner as Qwen or Llama models.
For instance, you can easily start a service using [vLLM](https://github.com/vllm-project/vllm):
```shell
vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --tensor-parallel-size 2 --max-model-len 32768 --enforce-eager
```
You can also easily start a service using [SGLang](https://github.com/sgl-project/sglang)
```bash
python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --trust-remote-code --tp 2
```
### Usage Recommendations
**We recommend adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:**
1. Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
2. **Avoid adding a system prompt; all instructions should be contained within the user prompt.**
3. For mathematical problems, it is advisable to include a directive in your prompt such as: "Please reason step by step, and put your final answer within \boxed{}."
4. When evaluating model performance, it is recommended to conduct multiple tests and average the results.
Additionally, we have observed that the DeepSeek-R1 series models tend to bypass thinking pattern (i.e., outputting "\<think\>\n\n\</think\>") when responding to certain queries, which can adversely affect the model's performance.
**To ensure that the model engages in thorough reasoning, we recommend enforcing the model to initiate its response with "\<think\>\n" at the beginning of every output.**
## 7. License
This code repository and the model weights are licensed under the [MIT License](https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE).
DeepSeek-R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs. Please note that:
- DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, DeepSeek-R1-Distill-Qwen-14B and DeepSeek-R1-Distill-Qwen-32B are derived from [Qwen-2.5 series](https://github.com/QwenLM/Qwen2.5), which are originally licensed under [Apache 2.0 License](https://huggingface.co/Qwen/Qwen2.5-1.5B/blob/main/LICENSE), and now finetuned with 800k samples curated with DeepSeek-R1.
- DeepSeek-R1-Distill-Llama-8B is derived from Llama3.1-8B-Base and is originally licensed under [llama3.1 license](https://huggingface.co/meta-llama/Llama-3.1-8B/blob/main/LICENSE).
- DeepSeek-R1-Distill-Llama-70B is derived from Llama3.3-70B-Instruct and is originally licensed under [llama3.3 license](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE).
## 8. Citation
```
@misc{deepseekai2025deepseekr1incentivizingreasoningcapability,
title={DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning},
author={DeepSeek-AI},
year={2025},
eprint={2501.12948},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2501.12948},
}
```
## 9. Contact
If you have any questions, please raise an issue or contact us at [[email protected]]([email protected]).
|
Mungert/rwkv7-0.4B-world-GGUF | Mungert | "2025-05-06T22:59:29Z" | 748 | 2 | null | [
"gguf",
"text-generation",
"en",
"zh",
"ja",
"ko",
"fr",
"ar",
"es",
"pt",
"base_model:BlinkDL/rwkv-7-world",
"base_model:quantized:BlinkDL/rwkv-7-world",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-18T09:56:18Z" | ---
license: apache-2.0
language:
- en
- zh
- ja
- ko
- fr
- ar
- es
- pt
metrics:
- accuracy
base_model:
- BlinkDL/rwkv-7-world
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">rwkv7-0.4B-world GGUF Models</span>
Note: you must use latest llama.cpp https://github.com/ggml-org/llama.cpp to run this model with llama.cpp
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `rwkv7-0.4B-world-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `rwkv7-0.4B-world-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `rwkv7-0.4B-world-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `rwkv7-0.4B-world-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `rwkv7-0.4B-world-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `rwkv7-0.4B-world-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `rwkv7-0.4B-world-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `rwkv7-0.4B-world-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `rwkv7-0.4B-world-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `rwkv7-0.4B-world-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `rwkv7-0.4B-world-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download) the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# rwkv7-0.4B-world
<!-- Provide a quick summary of what the model is/does. -->
This is RWKV-7 model under flash-linear attention format.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Bo Peng, Yu Zhang, Songlin Yang, Ruichong Zhang
- **Funded by:** RWKV Project (Under LF AI & Data Foundation)
- **Model type:** RWKV7
- **Language(s) (NLP):** English
- **License:** Apache-2.0
- **Parameter count:** 0.450B
- **Tokenizer:** RWKV World tokenizer
- **Vocabulary size:** 65,536
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** https://github.com/fla-org/flash-linear-attention ; https://github.com/BlinkDL/RWKV-LM
- **Paper:** With in Progress
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
Install `flash-linear-attention` and the latest version of `transformers` before using this model:
```bash
pip install git+https://github.com/fla-org/flash-linear-attention
pip install 'transformers>=4.48.0'
```
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
You can use this model just as any other HuggingFace models:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('fla-hub/rwkv7-0.4B-world', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('fla-hub/rwkv7-0.4B-world', trust_remote_code=True)
model = model.cuda()
prompt = "What is a large language model?"
messages = [
{"role": "user", "content": "Who are you?"},
{"role": "assistant", "content": "I am a GPT-3 based model."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=1024,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)[0]
print(response)
```
## Training Details
### Training Data
This model is trained on the World v3 with a total of 3.119 trillion tokens.
#### Training Hyperparameters
- **Training regime:** bfloat16, lr 4e-4 to 1e-5 "delayed" cosine decay, wd 0.1 (with increasing batch sizes during the middle)
## FAQ
Q: safetensors metadata is none.
A: upgrade transformers to >=4.48.0: `pip install 'transformers>=4.48.0'` |
Mungert/rwkv7-1.5B-world-GGUF | Mungert | "2025-05-06T22:59:26Z" | 558 | 1 | null | [
"gguf",
"text-generation",
"en",
"zh",
"ja",
"ko",
"fr",
"ar",
"es",
"pt",
"base_model:BlinkDL/rwkv-7-world",
"base_model:quantized:BlinkDL/rwkv-7-world",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-18T08:04:12Z" | ---
license: apache-2.0
language:
- en
- zh
- ja
- ko
- fr
- ar
- es
- pt
metrics:
- accuracy
base_model:
- BlinkDL/rwkv-7-world
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">rwkv7-1.5B-world GGUF Models</span>
Note: you must use latest llama.cpp https://github.com/ggml-org/llama.cpp to run this model with llama.cpp
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `rwkv7-1.5B-world-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `rwkv7-1.5B-world-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `rwkv7-1.5B-world-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `rwkv7-1.5B-world-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `rwkv7-1.5B-world-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `rwkv7-1.5B-world-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `rwkv7-1.5B-world-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `rwkv7-1.5B-world-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `rwkv7-1.5B-world-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `rwkv7-1.5B-world-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `rwkv7-1.5B-world-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download) the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# rwkv7-1.5B-world
<!-- Provide a quick summary of what the model is/does. -->
This is RWKV-7 model under flash-linear attention format.
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** Bo Peng, Yu Zhang, Songlin Yang, Ruichong Zhang
- **Funded by:** RWKV Project (Under LF AI & Data Foundation)
- **Model type:** RWKV7
- **Language(s) (NLP):** English
- **License:** Apache-2.0
- **Parameter count:** 1.52B
- **Tokenizer:** RWKV World tokenizer
- **Vocabulary size:** 65,536
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** https://github.com/fla-org/flash-linear-attention ; https://github.com/BlinkDL/RWKV-LM
- **Paper:** With in Progress
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
Install `flash-linear-attention` and the latest version of `transformers` before using this model:
```bash
pip install git+https://github.com/fla-org/flash-linear-attention
pip install 'transformers>=4.48.0'
```
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
You can use this model just as any other HuggingFace models:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('fla-hub/rwkv7-1.5B-world', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('fla-hub/rwkv7-1.5B-world', trust_remote_code=True)
model = model.cuda()
prompt = "What is a large language model?"
messages = [
{"role": "user", "content": "Who are you?"},
{"role": "assistant", "content": "I am a GPT-3 based model."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=1024,
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)[0]
print(response)
```
## Training Details
### Training Data
This model is trained on the World v3 with a total of 3.119 trillion tokens.
#### Training Hyperparameters
- **Training regime:** bfloat16, lr 4e-4 to 1e-5 "delayed" cosine decay, wd 0.1 (with increasing batch sizes during the middle)
- **Final Loss:** 1.9965
- **Token Count:** 3.119 trillion
## Evaluation
#### Metrics
`lambada_openai`:
before conversion: ppl 4.13 acc 69.4%
after conversion: ppl 4.26 acc 68.8% (without apply temple)
## FAQ
Q: safetensors metadata is none.
A: upgrade transformers to >=4.48.0: `pip install 'transformers>=4.48.0'` |
Mungert/TriLM_3.9B_Unpacked-GGUF | Mungert | "2025-05-06T22:59:10Z" | 4,789 | 3 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | "2025-03-17T11:16:59Z" | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_3.9B_Unpacked GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_3.9B_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_3.9B_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_3.9B_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_3.9B_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_3.9B_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_3.9B_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_3.9B_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_3.9B_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_3.9B_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_3.9B_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_3.9B_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# TriLM 3.9B Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_3.9B_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/Meta-Llama-3-8B-Instruct-GGUF | Mungert | "2025-05-06T22:59:06Z" | 1,372 | 3 | null | [
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"license:llama3",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-17T05:54:18Z" | ---
language:
- en
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3
new_version: meta-llama/Llama-3.1-8B-Instruct
extra_gated_prompt: >-
### META LLAMA 3 COMMUNITY LICENSE AGREEMENT
Meta Llama 3 Version Release Date: April 18, 2024
"Agreement" means the terms and conditions for use, reproduction, distribution and modification of the
Llama Materials set forth herein.
"Documentation" means the specifications, manuals and documentation accompanying Meta Llama 3
distributed by Meta at https://llama.meta.com/get-started/.
"Licensee" or "you" means you, or your employer or any other person or entity (if you are entering into
this Agreement on such person or entity’s behalf), of the age required under applicable laws, rules or
regulations to provide legal consent and that has legal authority to bind your employer or such other
person or entity if you are entering in this Agreement on their behalf.
"Meta Llama 3" means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://llama.meta.com/llama-downloads.
"Llama Materials" means, collectively, Meta’s proprietary Meta Llama 3 and Documentation (and any
portion thereof) made available under this Agreement.
"Meta" or "we" means Meta Platforms Ireland Limited (if you are located in or, if you are an entity, your
principal place of business is in the EEA or Switzerland) and Meta Platforms, Inc. (if you are located
outside of the EEA or Switzerland).
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide, non-transferable and royalty-free
limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama
Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the
Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works
thereof), or a product or service that uses any of them, including another AI model, you shall (A) provide
a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Meta
Llama 3” on a related website, user interface, blogpost, about page, or product documentation. If you
use the Llama Materials to create, train, fine tune, or otherwise improve an AI model, which is
distributed or made available, you shall also include “Llama 3” at the beginning of any such AI model
name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the following
attribution notice within a “Notice” text file distributed as a part of such copies: “Meta Llama 3 is
licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights
Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for the Llama
Materials (available at https://llama.meta.com/llama3/use-policy), which is hereby incorporated by
reference into this Agreement.
v. You will not use the Llama Materials or any output or results of the Llama Materials to
improve any other large language model (excluding Meta Llama 3 or derivative works thereof).
2. Additional Commercial Terms. If, on the Meta Llama 3 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700
million monthly active users in the preceding calendar month, you must request a license from Meta,
which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the
rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY
OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF
ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED,
INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT,
MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR
DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND
ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING
OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL,
INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN IF META OR ITS AFFILIATES HAVE BEEN ADVISED
OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama
Materials, neither Meta nor Licensee may use any name or mark owned by or associated with the other
or any of its affiliates, except as required for reasonable and customary use in describing and
redistributing the Llama Materials or as set forth in this Section 5(a). Meta hereby grants you a license to
use “Llama 3” (the “Mark”) solely as required to comply with the last sentence of Section 1.b.i. You will
comply with Meta’s brand guidelines (currently accessible at
https://about.meta.com/brand/resources/meta/company-brand/ ). All goodwill arising out of your use
of the Mark will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with
respect to any derivative works and modifications of the Llama Materials that are made by you, as
between you and Meta, you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Meta Llama 3 outputs or
results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other
rights owned or licensable by you, then any licenses granted to you under this Agreement shall
terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold
harmless Meta from and against any claim by any third party arising out of or related to your use or
distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this
Agreement or access to the Llama Materials and will continue in full force and effect until terminated in
accordance with the terms and conditions herein. Meta may terminate this Agreement if you are in
breach of any term or condition of this Agreement. Upon termination of this Agreement, you shall delete
and cease use of the Llama Materials. Sections 3, 4 and 7 shall survive the termination of this
Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of
the State of California without regard to choice of law principles, and the UN Convention on Contracts
for the International Sale of Goods does not apply to this Agreement. The courts of California shall have
exclusive jurisdiction of any dispute arising out of this Agreement.
### Meta Llama 3 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Meta Llama 3. If you
access or use Meta Llama 3, you agree to this Acceptable Use Policy (“Policy”). The most recent copy of
this policy can be found at [https://llama.meta.com/llama3/use-policy](https://llama.meta.com/llama3/use-policy)
#### Prohibited Uses
We want everyone to use Meta Llama 3 safely and responsibly. You agree you will not use, or allow
others to use, Meta Llama 3 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
2. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
3. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
4. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
5. Collect, process, disclose, generate, or infer health, demographic, or other sensitive personal or private information about individuals without rights and consents required by applicable laws
6. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
7. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Meta Llama 3 related to the following:
1. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State
2. Guns and illegal weapons (including weapon development)
3. Illegal drugs and regulated/controlled substances
4. Operation of critical infrastructure, transportation technologies, or heavy machinery
5. Self-harm or harm to others, including suicide, cutting, and eating disorders
6. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Meta Llama 3 related to the following:
1. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
2. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
3. Generating, promoting, or further distributing spam
4. Impersonating another individual without consent, authorization, or legal right
5. Representing that the use of Meta Llama 3 or outputs are human-generated
6. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation
of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama3](https://github.com/meta-llama/llama3)
* Reporting risky content generated by the model:
developers.facebook.com/llama_output_feedback
* Reporting bugs and security concerns: facebook.com/whitehat/info
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Meta Llama 3: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: The information you provide will be collected, stored, processed and shared in accordance with the [Meta Privacy Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
widget:
- example_title: Hello
messages:
- role: user
content: Hey my name is Julien! How are you?
- example_title: Winter holidays
messages:
- role: system
content: You are a helpful and honest assistant. Please, respond concisely and truthfully.
- role: user
content: Can you recommend a good destination for Winter holidays?
- example_title: Programming assistant
messages:
- role: system
content: You are a helpful and honest code and programming assistant. Please, respond concisely and truthfully.
- role: user
content: Write a function that computes the nth fibonacci number.
inference:
parameters:
max_new_tokens: 300
stop:
- <|end_of_text|>
- <|eot_id|>
---
# <span style="color: #7FFF7F;">Meta-Llama-3-8B-Instruct GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Meta-Llama-3-8B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Meta-Llama-3-8B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Meta-Llama-3-8B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Meta-Llama-3-8B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Meta-Llama-3-8B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Meta-Llama-3-8B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Meta-Llama-3-8B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Meta-Llama-3-8B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Meta-Llama-3-8B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Meta-Llama-3-8B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Meta-Llama-3-8B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
## Model Details
Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.
**Model developers** Meta
**Variations** Llama 3 comes in two sizes — 8B and 70B parameters — in pre-trained and instruction tuned variants.
**Input** Models input text only.
**Output** Models generate text and code only.
**Model Architecture** Llama 3 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
<table>
<tr>
<td>
</td>
<td><strong>Training Data</strong>
</td>
<td><strong>Params</strong>
</td>
<td><strong>Context length</strong>
</td>
<td><strong>GQA</strong>
</td>
<td><strong>Token count</strong>
</td>
<td><strong>Knowledge cutoff</strong>
</td>
</tr>
<tr>
<td rowspan="2" >Llama 3
</td>
<td rowspan="2" >A new mix of publicly available online data.
</td>
<td>8B
</td>
<td>8k
</td>
<td>Yes
</td>
<td rowspan="2" >15T+
</td>
<td>March, 2023
</td>
</tr>
<tr>
<td>70B
</td>
<td>8k
</td>
<td>Yes
</td>
<td>December, 2023
</td>
</tr>
</table>
**Llama 3 family of models**. Token counts refer to pretraining data only. Both the 8 and 70B versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date** April 18, 2024.
**Status** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.
**License** A custom commercial license is available at: [https://llama.meta.com/llama3/license](https://llama.meta.com/llama3/license)
Where to send questions or comments about the model Instructions on how to provide feedback or comments on the model can be found in the model [README](https://github.com/meta-llama/llama3). For more technical information about generation parameters and recipes for how to use Llama 3 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases** Llama 3 is intended for commercial and research use in English. Instruction tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.
**Out-of-scope** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3 Community License. Use in languages other than English**.
**Note: Developers may fine-tune Llama 3 models for languages beyond English provided they comply with the Llama 3 Community License and the Acceptable Use Policy.
## How to use
This repository contains two versions of Meta-Llama-3-8B-Instruct, for use with transformers and with the original `llama3` codebase.
### Use with transformers
You can run conversational inference using the Transformers pipeline abstraction, or by leveraging the Auto classes with the `generate()` function. Let's see examples of both.
#### Transformers pipeline
```python
import transformers
import torch
model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
terminators = [
pipeline.tokenizer.eos_token_id,
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = pipeline(
messages,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9,
)
print(outputs[0]["generated_text"][-1])
```
#### Transformers AutoModelForCausalLM
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
input_ids = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_tensors="pt"
).to(model.device)
terminators = [
tokenizer.eos_token_id,
tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = model.generate(
input_ids,
max_new_tokens=256,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```
### Use with `llama3`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama3)
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Meta-Llama-3-8B-Instruct --include "original/*" --local-dir Meta-Llama-3-8B-Instruct
```
For Hugging Face support, we recommend using transformers or TGI, but a similar command works.
## Hardware and Software
**Training Factors** We used custom training libraries, Meta's Research SuperCluster, and production clusters for pretraining. Fine-tuning, annotation, and evaluation were also performed on third-party cloud compute.
**Carbon Footprint Pretraining utilized a cumulative** 7.7M GPU hours of computation on hardware of type H100-80GB (TDP of 700W). Estimated total emissions were 2290 tCO2eq, 100% of which were offset by Meta’s sustainability program.
<table>
<tr>
<td>
</td>
<td><strong>Time (GPU hours)</strong>
</td>
<td><strong>Power Consumption (W)</strong>
</td>
<td><strong>Carbon Emitted(tCO2eq)</strong>
</td>
</tr>
<tr>
<td>Llama 3 8B
</td>
<td>1.3M
</td>
<td>700
</td>
<td>390
</td>
</tr>
<tr>
<td>Llama 3 70B
</td>
<td>6.4M
</td>
<td>700
</td>
<td>1900
</td>
</tr>
<tr>
<td>Total
</td>
<td>7.7M
</td>
<td>
</td>
<td>2290
</td>
</tr>
</table>
**CO2 emissions during pre-training**. Time: total GPU time required for training each model. Power Consumption: peak power capacity per GPU device for the GPUs used adjusted for power usage efficiency. 100% of the emissions are directly offset by Meta's sustainability program, and because we are openly releasing these models, the pretraining costs do not need to be incurred by others.
## Training Data
**Overview** Llama 3 was pretrained on over 15 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over 10M human-annotated examples. Neither the pretraining nor the fine-tuning datasets include Meta user data.
**Data Freshness** The pretraining data has a cutoff of March 2023 for the 8B and December 2023 for the 70B models respectively.
## Benchmarks
In this section, we report the results for Llama 3 models on standard automatic benchmarks. For all the evaluations, we use our internal evaluations library. For details on the methodology see [here](https://github.com/meta-llama/llama3/blob/main/eval_methodology.md).
### Base pretrained models
<table>
<tr>
<td><strong>Category</strong>
</td>
<td><strong>Benchmark</strong>
</td>
<td><strong>Llama 3 8B</strong>
</td>
<td><strong>Llama2 7B</strong>
</td>
<td><strong>Llama2 13B</strong>
</td>
<td><strong>Llama 3 70B</strong>
</td>
<td><strong>Llama2 70B</strong>
</td>
</tr>
<tr>
<td rowspan="6" >General
</td>
<td>MMLU (5-shot)
</td>
<td>66.6
</td>
<td>45.7
</td>
<td>53.8
</td>
<td>79.5
</td>
<td>69.7
</td>
</tr>
<tr>
<td>AGIEval English (3-5 shot)
</td>
<td>45.9
</td>
<td>28.8
</td>
<td>38.7
</td>
<td>63.0
</td>
<td>54.8
</td>
</tr>
<tr>
<td>CommonSenseQA (7-shot)
</td>
<td>72.6
</td>
<td>57.6
</td>
<td>67.6
</td>
<td>83.8
</td>
<td>78.7
</td>
</tr>
<tr>
<td>Winogrande (5-shot)
</td>
<td>76.1
</td>
<td>73.3
</td>
<td>75.4
</td>
<td>83.1
</td>
<td>81.8
</td>
</tr>
<tr>
<td>BIG-Bench Hard (3-shot, CoT)
</td>
<td>61.1
</td>
<td>38.1
</td>
<td>47.0
</td>
<td>81.3
</td>
<td>65.7
</td>
</tr>
<tr>
<td>ARC-Challenge (25-shot)
</td>
<td>78.6
</td>
<td>53.7
</td>
<td>67.6
</td>
<td>93.0
</td>
<td>85.3
</td>
</tr>
<tr>
<td>Knowledge reasoning
</td>
<td>TriviaQA-Wiki (5-shot)
</td>
<td>78.5
</td>
<td>72.1
</td>
<td>79.6
</td>
<td>89.7
</td>
<td>87.5
</td>
</tr>
<tr>
<td rowspan="4" >Reading comprehension
</td>
<td>SQuAD (1-shot)
</td>
<td>76.4
</td>
<td>72.2
</td>
<td>72.1
</td>
<td>85.6
</td>
<td>82.6
</td>
</tr>
<tr>
<td>QuAC (1-shot, F1)
</td>
<td>44.4
</td>
<td>39.6
</td>
<td>44.9
</td>
<td>51.1
</td>
<td>49.4
</td>
</tr>
<tr>
<td>BoolQ (0-shot)
</td>
<td>75.7
</td>
<td>65.5
</td>
<td>66.9
</td>
<td>79.0
</td>
<td>73.1
</td>
</tr>
<tr>
<td>DROP (3-shot, F1)
</td>
<td>58.4
</td>
<td>37.9
</td>
<td>49.8
</td>
<td>79.7
</td>
<td>70.2
</td>
</tr>
</table>
### Instruction tuned models
<table>
<tr>
<td><strong>Benchmark</strong>
</td>
<td><strong>Llama 3 8B</strong>
</td>
<td><strong>Llama 2 7B</strong>
</td>
<td><strong>Llama 2 13B</strong>
</td>
<td><strong>Llama 3 70B</strong>
</td>
<td><strong>Llama 2 70B</strong>
</td>
</tr>
<tr>
<td>MMLU (5-shot)
</td>
<td>68.4
</td>
<td>34.1
</td>
<td>47.8
</td>
<td>82.0
</td>
<td>52.9
</td>
</tr>
<tr>
<td>GPQA (0-shot)
</td>
<td>34.2
</td>
<td>21.7
</td>
<td>22.3
</td>
<td>39.5
</td>
<td>21.0
</td>
</tr>
<tr>
<td>HumanEval (0-shot)
</td>
<td>62.2
</td>
<td>7.9
</td>
<td>14.0
</td>
<td>81.7
</td>
<td>25.6
</td>
</tr>
<tr>
<td>GSM-8K (8-shot, CoT)
</td>
<td>79.6
</td>
<td>25.7
</td>
<td>77.4
</td>
<td>93.0
</td>
<td>57.5
</td>
</tr>
<tr>
<td>MATH (4-shot, CoT)
</td>
<td>30.0
</td>
<td>3.8
</td>
<td>6.7
</td>
<td>50.4
</td>
<td>11.6
</td>
</tr>
</table>
### Responsibility & Safety
We believe that an open approach to AI leads to better, safer products, faster innovation, and a bigger overall market. We are committed to Responsible AI development and took a series of steps to limit misuse and harm and support the open source community.
Foundation models are widely capable technologies that are built to be used for a diverse range of applications. They are not designed to meet every developer preference on safety levels for all use cases, out-of-the-box, as those by their nature will differ across different applications.
Rather, responsible LLM-application deployment is achieved by implementing a series of safety best practices throughout the development of such applications, from the model pre-training, fine-tuning and the deployment of systems composed of safeguards to tailor the safety needs specifically to the use case and audience.
As part of the Llama 3 release, we updated our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/) to outline the steps and best practices for developers to implement model and system level safety for their application. We also provide a set of resources including [Meta Llama Guard 2](https://llama.meta.com/purple-llama/) and [Code Shield](https://llama.meta.com/purple-llama/) safeguards. These tools have proven to drastically reduce residual risks of LLM Systems, while maintaining a high level of helpfulness. We encourage developers to tune and deploy these safeguards according to their needs and we provide a [reference implementation](https://github.com/meta-llama/llama-recipes/tree/main/recipes/responsible_ai) to get you started.
#### Llama 3-Instruct
As outlined in the Responsible Use Guide, some trade-off between model helpfulness and model alignment is likely unavoidable. Developers should exercise discretion about how to weigh the benefits of alignment and helpfulness for their specific use case and audience. Developers should be mindful of residual risks when using Llama models and leverage additional safety tools as needed to reach the right safety bar for their use case.
<span style="text-decoration:underline;">Safety</span>
For our instruction tuned model, we conducted extensive red teaming exercises, performed adversarial evaluations and implemented safety mitigations techniques to lower residual risks. As with any Large Language Model, residual risks will likely remain and we recommend that developers assess these risks in the context of their use case. In parallel, we are working with the community to make AI safety benchmark standards transparent, rigorous and interpretable.
<span style="text-decoration:underline;">Refusals</span>
In addition to residual risks, we put a great emphasis on model refusals to benign prompts. Over-refusing not only can impact the user experience but could even be harmful in certain contexts as well. We’ve heard the feedback from the developer community and improved our fine tuning to ensure that Llama 3 is significantly less likely to falsely refuse to answer prompts than Llama 2.
We built internal benchmarks and developed mitigations to limit false refusals making Llama 3 our most helpful model to date.
#### Responsible release
In addition to responsible use considerations outlined above, we followed a rigorous process that requires us to take extra measures against misuse and critical risks before we make our release decision.
Misuse
If you access or use Llama 3, you agree to the Acceptable Use Policy. The most recent copy of this policy can be found at [https://llama.meta.com/llama3/use-policy/](https://llama.meta.com/llama3/use-policy/).
#### Critical risks
<span style="text-decoration:underline;">CBRNE</span> (Chemical, Biological, Radiological, Nuclear, and high yield Explosives)
We have conducted a two fold assessment of the safety of the model in this area:
* Iterative testing during model training to assess the safety of responses related to CBRNE threats and other adversarial risks.
* Involving external CBRNE experts to conduct an uplift test assessing the ability of the model to accurately provide expert knowledge and reduce barriers to potential CBRNE misuse, by reference to what can be achieved using web search (without the model).
### <span style="text-decoration:underline;">Cyber Security </span>
We have evaluated Llama 3 with CyberSecEval, Meta’s cybersecurity safety eval suite, measuring Llama 3’s propensity to suggest insecure code when used as a coding assistant, and Llama 3’s propensity to comply with requests to help carry out cyber attacks, where attacks are defined by the industry standard MITRE ATT&CK cyber attack ontology. On our insecure coding and cyber attacker helpfulness tests, Llama 3 behaved in the same range or safer than models of [equivalent coding capability](https://huggingface.co/spaces/facebook/CyberSecEval).
### <span style="text-decoration:underline;">Child Safety</span>
Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
### Community
Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership in AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
The core values of Llama 3 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
But Llama 3 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has been in English, and has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3 models, developers should perform safety testing and tuning tailored to their specific applications of the model. As outlined in the Responsible Use Guide, we recommend incorporating [Purple Llama](https://github.com/facebookresearch/PurpleLlama) solutions into your workflows and specifically [Llama Guard](https://ai.meta.com/research/publications/llama-guard-llm-based-input-output-safeguard-for-human-ai-conversations/) which provides a base model to filter input and output prompts to layer system-level safety on top of model-level safety.
Please see the Responsible Use Guide available at [http://llama.meta.com/responsible-use-guide](http://llama.meta.com/responsible-use-guide)
## Citation instructions
@article{llama3modelcard,
title={Llama 3 Model Card},
author={AI@Meta},
year={2024},
url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
}
## Contributors
Aaditya Singh; Aaron Grattafiori; Abhimanyu Dubey; Abhinav Jauhri; Abhinav Pandey; Abhishek Kadian; Adam Kelsey; Adi Gangidi; Ahmad Al-Dahle; Ahuva Goldstand; Aiesha Letman; Ajay Menon; Akhil Mathur; Alan Schelten; Alex Vaughan; Amy Yang; Andrei Lupu; Andres Alvarado; Andrew Gallagher; Andrew Gu; Andrew Ho; Andrew Poulton; Andrew Ryan; Angela Fan; Ankit Ramchandani; Anthony Hartshorn; Archi Mitra; Archie Sravankumar; Artem Korenev; Arun Rao; Ashley Gabriel; Ashwin Bharambe; Assaf Eisenman; Aston Zhang; Aurelien Rodriguez; Austen Gregerson; Ava Spataru; Baptiste Roziere; Ben Maurer; Benjamin Leonhardi; Bernie Huang; Bhargavi Paranjape; Bing Liu; Binh Tang; Bobbie Chern; Brani Stojkovic; Brian Fuller; Catalina Mejia Arenas; Chao Zhou; Charlotte Caucheteux; Chaya Nayak; Ching-Hsiang Chu; Chloe Bi; Chris Cai; Chris Cox; Chris Marra; Chris McConnell; Christian Keller; Christoph Feichtenhofer; Christophe Touret; Chunyang Wu; Corinne Wong; Cristian Canton Ferrer; Damien Allonsius; Daniel Kreymer; Daniel Haziza; Daniel Li; Danielle Pintz; Danny Livshits; Danny Wyatt; David Adkins; David Esiobu; David Xu; Davide Testuggine; Delia David; Devi Parikh; Dhruv Choudhary; Dhruv Mahajan; Diana Liskovich; Diego Garcia-Olano; Diego Perino; Dieuwke Hupkes; Dingkang Wang; Dustin Holland; Egor Lakomkin; Elina Lobanova; Xiaoqing Ellen Tan; Emily Dinan; Eric Smith; Erik Brinkman; Esteban Arcaute; Filip Radenovic; Firat Ozgenel; Francesco Caggioni; Frank Seide; Frank Zhang; Gabriel Synnaeve; Gabriella Schwarz; Gabrielle Lee; Gada Badeer; Georgia Anderson; Graeme Nail; Gregoire Mialon; Guan Pang; Guillem Cucurell; Hailey Nguyen; Hannah Korevaar; Hannah Wang; Haroun Habeeb; Harrison Rudolph; Henry Aspegren; Hu Xu; Hugo Touvron; Iga Kozlowska; Igor Molybog; Igor Tufanov; Iliyan Zarov; Imanol Arrieta Ibarra; Irina-Elena Veliche; Isabel Kloumann; Ishan Misra; Ivan Evtimov; Jacob Xu; Jade Copet; Jake Weissman; Jan Geffert; Jana Vranes; Japhet Asher; Jason Park; Jay Mahadeokar; Jean-Baptiste Gaya; Jeet Shah; Jelmer van der Linde; Jennifer Chan; Jenny Hong; Jenya Lee; Jeremy Fu; Jeremy Teboul; Jianfeng Chi; Jianyu Huang; Jie Wang; Jiecao Yu; Joanna Bitton; Joe Spisak; Joelle Pineau; Jon Carvill; Jongsoo Park; Joseph Rocca; Joshua Johnstun; Junteng Jia; Kalyan Vasuden Alwala; Kam Hou U; Kate Plawiak; Kartikeya Upasani; Kaushik Veeraraghavan; Ke Li; Kenneth Heafield; Kevin Stone; Khalid El-Arini; Krithika Iyer; Kshitiz Malik; Kuenley Chiu; Kunal Bhalla; Kyle Huang; Lakshya Garg; Lauren Rantala-Yeary; Laurens van der Maaten; Lawrence Chen; Leandro Silva; Lee Bell; Lei Zhang; Liang Tan; Louis Martin; Lovish Madaan; Luca Wehrstedt; Lukas Blecher; Luke de Oliveira; Madeline Muzzi; Madian Khabsa; Manav Avlani; Mannat Singh; Manohar Paluri; Mark Zuckerberg; Marcin Kardas; Martynas Mankus; Mathew Oldham; Mathieu Rita; Matthew Lennie; Maya Pavlova; Meghan Keneally; Melanie Kambadur; Mihir Patel; Mikayel Samvelyan; Mike Clark; Mike Lewis; Min Si; Mitesh Kumar Singh; Mo Metanat; Mona Hassan; Naman Goyal; Narjes Torabi; Nicolas Usunier; Nikolay Bashlykov; Nikolay Bogoychev; Niladri Chatterji; Ning Dong; Oliver Aobo Yang; Olivier Duchenne; Onur Celebi; Parth Parekh; Patrick Alrassy; Paul Saab; Pavan Balaji; Pedro Rittner; Pengchuan Zhang; Pengwei Li; Petar Vasic; Peter Weng; Polina Zvyagina; Prajjwal Bhargava; Pratik Dubal; Praveen Krishnan; Punit Singh Koura; Qing He; Rachel Rodriguez; Ragavan Srinivasan; Rahul Mitra; Ramon Calderer; Raymond Li; Robert Stojnic; Roberta Raileanu; Robin Battey; Rocky Wang; Rohit Girdhar; Rohit Patel; Romain Sauvestre; Ronnie Polidoro; Roshan Sumbaly; Ross Taylor; Ruan Silva; Rui Hou; Rui Wang; Russ Howes; Ruty Rinott; Saghar Hosseini; Sai Jayesh Bondu; Samyak Datta; Sanjay Singh; Sara Chugh; Sargun Dhillon; Satadru Pan; Sean Bell; Sergey Edunov; Shaoliang Nie; Sharan Narang; Sharath Raparthy; Shaun Lindsay; Sheng Feng; Sheng Shen; Shenghao Lin; Shiva Shankar; Shruti Bhosale; Shun Zhang; Simon Vandenhende; Sinong Wang; Seohyun Sonia Kim; Soumya Batra; Sten Sootla; Steve Kehoe; Suchin Gururangan; Sumit Gupta; Sunny Virk; Sydney Borodinsky; Tamar Glaser; Tamar Herman; Tamara Best; Tara Fowler; Thomas Georgiou; Thomas Scialom; Tianhe Li; Todor Mihaylov; Tong Xiao; Ujjwal Karn; Vedanuj Goswami; Vibhor Gupta; Vignesh Ramanathan; Viktor Kerkez; Vinay Satish Kumar; Vincent Gonguet; Vish Vogeti; Vlad Poenaru; Vlad Tiberiu Mihailescu; Vladan Petrovic; Vladimir Ivanov; Wei Li; Weiwei Chu; Wenhan Xiong; Wenyin Fu; Wes Bouaziz; Whitney Meers; Will Constable; Xavier Martinet; Xiaojian Wu; Xinbo Gao; Xinfeng Xie; Xuchao Jia; Yaelle Goldschlag; Yann LeCun; Yashesh Gaur; Yasmine Babaei; Ye Qi; Yenda Li; Yi Wen; Yiwen Song; Youngjin Nam; Yuchen Hao; Yuchen Zhang; Yun Wang; Yuning Mao; Yuzi He; Zacharie Delpierre Coudert; Zachary DeVito; Zahra Hankir; Zhaoduo Wen; Zheng Yan; Zhengxing Chen; Zhenyu Yang; Zoe Papakipos
|
Mungert/TriLM_1.1B_Unpacked-GGUF | Mungert | "2025-05-06T22:59:02Z" | 441 | 0 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | "2025-03-17T03:04:19Z" | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_1.1B_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_1.1B_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_1.1B_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_1.1B_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_1.1B_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_1.1B_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_1.1B_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_1.1B_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_1.1B_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_1.1B_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_1.1B_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_1.1B_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# TriLM 1.1B Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_1.1B_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
ma921/phi2_h_dpo_oasst1_noise40_epoch3 | ma921 | "2025-05-06T22:58:50Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"phi",
"text-generation",
"generated_from_trainer",
"base_model:ma921/phi-2-sft-oasst1",
"base_model:finetune:ma921/phi-2-sft-oasst1",
"license:mit",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-06T22:55:02Z" | ---
library_name: transformers
license: mit
base_model: ma921/phi-2-sft-oasst1
tags:
- generated_from_trainer
model-index:
- name: phi2_h_dpo_oasst1_noise40_epoch3
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# phi2_h_dpo_oasst1_noise40_epoch3
This model is a fine-tuned version of [ma921/phi-2-sft-oasst1](https://huggingface.co/ma921/phi-2-sft-oasst1) on an unknown dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-06
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 128
- total_train_batch_size: 256
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 2
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.51.3
- Pytorch 2.6.0+cu124
- Datasets 3.5.1
- Tokenizers 0.21.1
|
Mungert/TriLM_190M_Unpacked-GGUF | Mungert | "2025-05-06T22:58:48Z" | 338 | 1 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | "2025-03-16T22:34:44Z" | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_190M_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_190M_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_190M_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_190M_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_190M_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_190M_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_190M_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_190M_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_190M_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_190M_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_190M_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_190M_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download) the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# TriLM 190M Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_190M_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/TriLM_99M_Unpacked-GGUF | Mungert | "2025-05-06T22:58:47Z" | 308 | 0 | null | [
"gguf",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | null | "2025-03-16T21:55:35Z" | ---
license: apache-2.0
---
# <span style="color: #7FFF7F;">TriLM_99M_Unpacked GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device’s specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn’t available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `TriLM_99M_Unpacked-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `TriLM_99M_Unpacked-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `TriLM_99M_Unpacked-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `TriLM_99M_Unpacked-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `TriLM_99M_Unpacked-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `TriLM_99M_Unpacked-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `TriLM_99M_Unpacked-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `TriLM_99M_Unpacked-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `TriLM_99M_Unpacked-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `TriLM_99M_Unpacked-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `TriLM_99M_Unpacked-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 [Network Monitor Assitant](https://readyforquantum.com).
💬 Click the **chat icon** (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.
### What I'm Testing
I'm experimenting with **function calling** against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".
🟡 **TestLLM** – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .
### The other Available AI Assistants
🟢 **TurboLLM** – Uses **gpt-4o-mini** Fast! . Note: tokens are limited since OpenAI models are pricey, but you can [Login](https://readyforquantum.com) or [Download](https://readyforquantum.com/download) the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .
🔵 **HugLLM** – Runs **open-source Hugging Face models** Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)
# TriLM 99M Unpacked
TriLM (ternary model), unpacked to FP16 format - compatible with FP16 GEMMs. After unpacking, TriLM has the same architecture as LLaMa.
```python
import transformers as tf, torch
model_name = "SpectraSuite/TriLM_99M_Unpacked"
# Please adjust the temperature, repetition penalty, top_k, top_p and other sampling parameters according to your needs.
pipeline = tf.pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.float16}, device_map="auto")
# These are base (pretrained) LLMs that are not instruction and chat tuned. You may need to adjust your prompt accordingly.
pipeline("Once upon a time")
```
* License: Apache 2.0
* We will use our GitHub repo for communication (including HF repo related queries). Feel free to open an issue here https://github.com/NolanoOrg/SpectraSuite
|
Mungert/Mistral-7B-Instruct-v0.2-GGUF | Mungert | "2025-05-06T22:58:41Z" | 481 | 2 | null | [
"gguf",
"finetuned",
"text-generation",
"arxiv:2310.06825",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-16T17:56:11Z" | ---
license: apache-2.0
tags:
- finetuned
pipeline_tag: text-generation
new_version: mistralai/Mistral-7B-Instruct-v0.3
inference: true
widget:
- messages:
- role: user
content: What is your favorite condiment?
extra_gated_description: If you want to learn more about how we process your personal data, please read our <a href="https://mistral.ai/terms/">Privacy Policy</a>.
---
# <span style="color: #7FFF7F;">Mistral-7B-Instruct-v0.2 GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Mistral-7B-Instruct-v0.2-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Mistral-7B-Instruct-v0.2-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Mistral-7B-Instruct-v0.2-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Mistral-7B-Instruct-v0.2-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Mistral-7B-Instruct-v0.2-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Mistral-7B-Instruct-v0.2-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Mistral-7B-Instruct-v0.2-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Mistral-7B-Instruct-v0.2-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Mistral-7B-Instruct-v0.2-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Mistral-7B-Instruct-v0.2-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Mistral-7B-Instruct-v0.2-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Model Card for Mistral-7B-Instruct-v0.2
## Encode and Decode with `mistral_common`
```py
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.protocol.instruct.messages import UserMessage
from mistral_common.protocol.instruct.request import ChatCompletionRequest
mistral_models_path = "MISTRAL_MODELS_PATH"
tokenizer = MistralTokenizer.v1()
completion_request = ChatCompletionRequest(messages=[UserMessage(content="Explain Machine Learning to me in a nutshell.")])
tokens = tokenizer.encode_chat_completion(completion_request).tokens
```
## Inference with `mistral_inference`
```py
from mistral_inference.transformer import Transformer
from mistral_inference.generate import generate
model = Transformer.from_folder(mistral_models_path)
out_tokens, _ = generate([tokens], model, max_tokens=64, temperature=0.0, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.decode(out_tokens[0])
print(result)
```
## Inference with hugging face `transformers`
```py
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
model.to("cuda")
generated_ids = model.generate(tokens, max_new_tokens=1000, do_sample=True)
# decode with mistral tokenizer
result = tokenizer.decode(generated_ids[0].tolist())
print(result)
```
> [!TIP]
> PRs to correct the `transformers` tokenizer so that it gives 1-to-1 the same results as the `mistral_common` reference implementation are very welcome!
---
The Mistral-7B-Instruct-v0.2 Large Language Model (LLM) is an instruct fine-tuned version of the Mistral-7B-v0.2.
Mistral-7B-v0.2 has the following changes compared to Mistral-7B-v0.1
- 32k context window (vs 8k context in v0.1)
- Rope-theta = 1e6
- No Sliding-Window Attention
For full details of this model please read our [paper](https://arxiv.org/abs/2310.06825) and [release blog post](https://mistral.ai/news/la-plateforme/).
## Instruction format
In order to leverage instruction fine-tuning, your prompt should be surrounded by `[INST]` and `[/INST]` tokens. The very first instruction should begin with a begin of sentence id. The next instructions should not. The assistant generation will be ended by the end-of-sentence token id.
E.g.
```
text = "<s>[INST] What is your favourite condiment? [/INST]"
"Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s> "
"[INST] Do you have mayonnaise recipes? [/INST]"
```
This format is available as a [chat template](https://huggingface.co/docs/transformers/main/chat_templating) via the `apply_chat_template()` method:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
messages = [
{"role": "user", "content": "What is your favourite condiment?"},
{"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
{"role": "user", "content": "Do you have mayonnaise recipes?"}
]
encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt")
model_inputs = encodeds.to(device)
model.to(device)
generated_ids = model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])
```
## Troubleshooting
- If you see the following error:
```
Traceback (most recent call last):
File "", line 1, in
File "/transformers/models/auto/auto_factory.py", line 482, in from_pretrained
config, kwargs = AutoConfig.from_pretrained(
File "/transformers/models/auto/configuration_auto.py", line 1022, in from_pretrained
config_class = CONFIG_MAPPING[config_dict["model_type"]]
File "/transformers/models/auto/configuration_auto.py", line 723, in getitem
raise KeyError(key)
KeyError: 'mistral'
```
Installing transformers from source should solve the issue
pip install git+https://github.com/huggingface/transformers
This should not be required after transformers-v4.33.4.
## Limitations
The Mistral 7B Instruct model is a quick demonstration that the base model can be easily fine-tuned to achieve compelling performance.
It does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to
make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
## The Mistral AI Team
Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Louis Ternon, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Théophile Gervet, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed. |
Mungert/OlympicCoder-7B-GGUF | Mungert | "2025-05-06T22:58:36Z" | 914 | 4 | transformers | [
"transformers",
"gguf",
"text-generation",
"en",
"dataset:open-r1/codeforces-cots",
"base_model:Qwen/Qwen2.5-Coder-7B-Instruct",
"base_model:quantized:Qwen/Qwen2.5-Coder-7B-Instruct",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-16T02:40:47Z" | ---
license: apache-2.0
datasets:
- open-r1/codeforces-cots
language:
- en
base_model:
- Qwen/Qwen2.5-Coder-7B-Instruct
pipeline_tag: text-generation
library_name: transformers
---
# <span style="color: #7FFF7F;">OlympicCoder-7B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `OlympicCoder-7B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `OlympicCoder-7B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `OlympicCoder-7B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `OlympicCoder-7B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `OlympicCoder-7B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `OlympicCoder-7B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `OlympicCoder-7B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `OlympicCoder-7B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `OlympicCoder-7B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `OlympicCoder-7B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `OlympicCoder-7B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Model Card for OlympicCoder-7B
OlympicCoder-7B is a code model that achieves strong performance on competitive coding benchmarks such as LiveCodeBench and the 2024 International Olympiad in Informatics.
* Repository: https://github.com/huggingface/open-r1
* Blog post: https://huggingface.co/blog/open-r1/update-3
## Model description
- **Model type:** A 7B parameter model fine-tuned on a decontaminated version of the codeforces dataset.
- **Language(s) (NLP):** Primarily English
- **License:** apache-2.0
- **Finetuned from model:** [Qwen/Qwen2.5-Coder-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct)
## Evaluation
We compare the performance of OlympicCoder models on two main benchmarks for competitive coding:
* **[IOI'2024:](https://github.com/huggingface/ioi)** 6 very challenging problems from the 2024 International Olympiad in Informatics. Models are allowed up to 50 submissions per problem.
* **[LiveCodeBench:](https://livecodebench.github.io)** Python programming problems source from platforms like CodeForces and LeetCoder. We use the `v4_v5` subset of [`livecodebench/code_generation_lite`](https://huggingface.co/datasets/livecodebench/code_generation_lite), which corresponds to 268 problems. We use `lighteval` to evaluate models on LiveCodeBench using the sampling parameters described [here](https://github.com/huggingface/open-r1?tab=readme-ov-file#livecodebench).
> [!NOTE]
> The OlympicCoder models were post-trained exclusively on C++ solutions generated by DeepSeek-R1. As a result the performance on LiveCodeBench should be considered to be partially _out-of-domain_, since this expects models to output solutions in Python.
### IOI'24
### LiveCodeBench
## Usage
Here's how you can run the model using the `pipeline()` function from 🤗 Transformers:
```python
# pip install transformers
# pip install accelerate
import torch
from transformers import pipeline
pipe = pipeline("text-generation", model="open-r1/OlympicCoder-7B", torch_dtype=torch.bfloat16, device_map="auto")
# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating
messages = [
{"role": "user", "content": "Write a python program to calculate the 10th Fibonacci number"},
]
prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, max_new_tokens=8000, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
#<|im_start|>user
#Write a python program to calculate the 10th fibonacci number<|im_end|>
#<|im_start|>assistant
#<think>Okay, I need to write a Python program that calculates the 10th Fibonacci number. Hmm, the Fibonacci sequence starts with 0 and 1. Each subsequent number is the sum of the two preceding ones. So the sequence goes: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, and so on. ...
```
> [!WARNING]
> To ensure that the model consistently outputs a long chain-of-thought, we have edited the chat template to prefill the first assistant turn with a `<think>` token. As a result, the outputs from this model will not show the opening `<think>` token if you use the model's `generate()` method. To apply reinforcement learning with a format reward, either prepend the `<think>` token to the model's completions or amend the chat template to remove the prefill.
## Training procedure
### Training hyper-parameters
The following hyperparameters were used during training:
- dataset: open-r1/codeforces-cots
- learning_rate: 4.0e-5
- train_batch_size: 2
- seed: 42
- packing: false
- distributed_type: deepspeed-zero-3
- num_devices: 8
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine_with_min_lr
- min_lr_rate: 0.1
- lr_scheduler_warmup_ratio: 0.03
- num_epochs: 10.0 |
Mungert/Llama-3.2-1B-Instruct-GGUF | Mungert | "2025-05-06T22:58:15Z" | 806 | 3 | transformers | [
"transformers",
"gguf",
"facebook",
"meta",
"pytorch",
"llama",
"llama-3",
"text-generation",
"en",
"de",
"fr",
"it",
"pt",
"hi",
"es",
"th",
"arxiv:2204.05149",
"arxiv:2405.16406",
"license:llama3.2",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-15T03:34:52Z" | ---
language:
- en
- de
- fr
- it
- pt
- hi
- es
- th
library_name: transformers
pipeline_tag: text-generation
tags:
- facebook
- meta
- pytorch
- llama
- llama-3
license: llama3.2
extra_gated_prompt: >-
### LLAMA 3.2 COMMUNITY LICENSE AGREEMENT
Llama 3.2 Version Release Date: September 25, 2024
“Agreement” means the terms and conditions for use, reproduction, distribution
and modification of the Llama Materials set forth herein.
“Documentation” means the specifications, manuals and documentation accompanying Llama 3.2
distributed by Meta at https://llama.meta.com/doc/overview.
“Licensee” or “you” means you, or your employer or any other person or entity (if you are
entering into this Agreement on such person or entity’s behalf), of the age required under
applicable laws, rules or regulations to provide legal consent and that has legal authority
to bind your employer or such other person or entity if you are entering in this Agreement
on their behalf.
“Llama 3.2” means the foundational large language models and software and algorithms, including
machine-learning model code, trained model weights, inference-enabling code, training-enabling code,
fine-tuning enabling code and other elements of the foregoing distributed by Meta at
https://www.llama.com/llama-downloads.
“Llama Materials” means, collectively, Meta’s proprietary Llama 3.2 and Documentation (and
any portion thereof) made available under this Agreement.
“Meta” or “we” means Meta Platforms Ireland Limited (if you are located in or,
if you are an entity, your principal place of business is in the EEA or Switzerland)
and Meta Platforms, Inc. (if you are located outside of the EEA or Switzerland).
By clicking “I Accept” below or by using or distributing any portion or element of the Llama Materials,
you agree to be bound by this Agreement.
1. License Rights and Redistribution.
a. Grant of Rights. You are granted a non-exclusive, worldwide,
non-transferable and royalty-free limited license under Meta’s intellectual property or other rights
owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works
of, and make modifications to the Llama Materials.
b. Redistribution and Use.
i. If you distribute or make available the Llama Materials (or any derivative works thereof),
or a product or service (including another AI model) that contains any of them, you shall (A) provide
a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Llama”
on a related website, user interface, blogpost, about page, or product documentation. If you use the
Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or
otherwise improve an AI model, which is distributed or made available, you shall also include “Llama”
at the beginning of any such AI model name.
ii. If you receive Llama Materials, or any derivative works thereof, from a Licensee as part
of an integrated end user product, then Section 2 of this Agreement will not apply to you.
iii. You must retain in all copies of the Llama Materials that you distribute the
following attribution notice within a “Notice” text file distributed as a part of such copies:
“Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms,
Inc. All Rights Reserved.”
iv. Your use of the Llama Materials must comply with applicable laws and regulations
(including trade compliance laws and regulations) and adhere to the Acceptable Use Policy for
the Llama Materials (available at https://www.llama.com/llama3_2/use-policy), which is hereby
incorporated by reference into this Agreement.
2. Additional Commercial Terms. If, on the Llama 3.2 version release date, the monthly active users
of the products or services made available by or for Licensee, or Licensee’s affiliates,
is greater than 700 million monthly active users in the preceding calendar month, you must request
a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to
exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND
RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS
ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES
OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE
FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED
WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.
4. Limitation of Liability. IN NO EVENT WILL META OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF LIABILITY,
WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT,
FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN
IF META OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY OF ANY OF THE FOREGOING.
5. Intellectual Property.
a. No trademark licenses are granted under this Agreement, and in connection with the Llama Materials,
neither Meta nor Licensee may use any name or mark owned by or associated with the other or any of its affiliates,
except as required for reasonable and customary use in describing and redistributing the Llama Materials or as
set forth in this Section 5(a). Meta hereby grants you a license to use “Llama” (the “Mark”) solely as required
to comply with the last sentence of Section 1.b.i. You will comply with Meta’s brand guidelines (currently accessible
at https://about.meta.com/brand/resources/meta/company-brand/). All goodwill arising out of your use of the Mark
will inure to the benefit of Meta.
b. Subject to Meta’s ownership of Llama Materials and derivatives made by or for Meta, with respect to any
derivative works and modifications of the Llama Materials that are made by you, as between you and Meta,
you are and will be the owner of such derivative works and modifications.
c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or
counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.2 outputs or results, or any portion
of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable
by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or
claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third
party arising out of or related to your use or distribution of the Llama Materials.
6. Term and Termination. The term of this Agreement will commence upon your acceptance of this Agreement or access
to the Llama Materials and will continue in full force and effect until terminated in accordance with the terms
and conditions herein. Meta may terminate this Agreement if you are in breach of any term or condition of this
Agreement. Upon termination of this Agreement, you shall delete and cease use of the Llama Materials. Sections 3,
4 and 7 shall survive the termination of this Agreement.
7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of the State of
California without regard to choice of law principles, and the UN Convention on Contracts for the International
Sale of Goods does not apply to this Agreement. The courts of California shall have exclusive jurisdiction of
any dispute arising out of this Agreement.
### Llama 3.2 Acceptable Use Policy
Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.2.
If you access or use Llama 3.2, you agree to this Acceptable Use Policy (“**Policy**”).
The most recent copy of this policy can be found at
[https://www.llama.com/llama3_2/use-policy](https://www.llama.com/llama3_2/use-policy).
#### Prohibited Uses
We want everyone to use Llama 3.2 safely and responsibly. You agree you will not use, or allow others to use, Llama 3.2 to:
1. Violate the law or others’ rights, including to:
1. Engage in, promote, generate, contribute to, encourage, plan, incite, or further illegal or unlawful activity or content, such as:
1. Violence or terrorism
2. Exploitation or harm to children, including the solicitation, creation, acquisition, or dissemination of child exploitative content or failure to report Child Sexual Abuse Material
3. Human trafficking, exploitation, and sexual violence
4. The illegal distribution of information or materials to minors, including obscene materials, or failure to employ legally required age-gating in connection with such information or materials.
5. Sexual solicitation
6. Any other criminal activity
1. Engage in, promote, incite, or facilitate the harassment, abuse, threatening, or bullying of individuals or groups of individuals
2. Engage in, promote, incite, or facilitate discrimination or other unlawful or harmful conduct in the provision of employment, employment benefits, credit, housing, other economic benefits, or other essential goods and services
3. Engage in the unauthorized or unlicensed practice of any profession including, but not limited to, financial, legal, medical/health, or related professional practices
4. Collect, process, disclose, generate, or infer private or sensitive information about individuals, including information about individuals’ identity, health, or demographic information, unless you have obtained the right to do so in accordance with applicable law
5. Engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the Llama Materials
6. Create, generate, or facilitate the creation of malicious code, malware, computer viruses or do anything else that could disable, overburden, interfere with or impair the proper working, integrity, operation or appearance of a website or computer system
7. Engage in any action, or facilitate any action, to intentionally circumvent or remove usage restrictions or other safety measures, or to enable functionality disabled by Meta
2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.2 related to the following:
8. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State or to the U.S. Biological Weapons Anti-Terrorism Act of 1989 or the Chemical Weapons Convention Implementation Act of 1997
9. Guns and illegal weapons (including weapon development)
10. Illegal drugs and regulated/controlled substances
11. Operation of critical infrastructure, transportation technologies, or heavy machinery
12. Self-harm or harm to others, including suicide, cutting, and eating disorders
13. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual
3. Intentionally deceive or mislead others, including use of Llama 3.2 related to the following:
14. Generating, promoting, or furthering fraud or the creation or promotion of disinformation
15. Generating, promoting, or furthering defamatory content, including the creation of defamatory statements, images, or other content
16. Generating, promoting, or further distributing spam
17. Impersonating another individual without consent, authorization, or legal right
18. Representing that the use of Llama 3.2 or outputs are human-generated
19. Generating or facilitating false online engagement, including fake reviews and other means of fake online engagement
4. Fail to appropriately disclose to end users any known dangers of your AI system
5. Interact with third party tools, models, or software designed to generate unlawful content or engage in unlawful or harmful conduct and/or represent that the outputs of such tools, models, or software are associated with Meta or Llama 3.2
With respect to any multimodal models included in Llama 3.2, the rights granted under Section 1(a) of the Llama 3.2 Community License Agreement are not being granted to you if you are an individual domiciled in, or a company with a principal place of business in, the European Union. This restriction does not apply to end users of a product or service that incorporates any such multimodal models.
Please report any violation of this Policy, software “bug,” or other problems that could lead to a violation of this Policy through one of the following means:
* Reporting issues with the model: [https://github.com/meta-llama/llama-models/issues](https://l.workplace.com/l.php?u=https%3A%2F%2Fgithub.com%2Fmeta-llama%2Fllama-models%2Fissues&h=AT0qV8W9BFT6NwihiOHRuKYQM_UnkzN_NmHMy91OT55gkLpgi4kQupHUl0ssR4dQsIQ8n3tfd0vtkobvsEvt1l4Ic6GXI2EeuHV8N08OG2WnbAmm0FL4ObkazC6G_256vN0lN9DsykCvCqGZ)
* Reporting risky content generated by the model: [developers.facebook.com/llama_output_feedback](http://developers.facebook.com/llama_output_feedback)
* Reporting bugs and security concerns: [facebook.com/whitehat/info](http://facebook.com/whitehat/info)
* Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama 3.2: [email protected]
extra_gated_fields:
First Name: text
Last Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Job title:
type: select
options:
- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
geo: ip_location
By clicking Submit below I accept the terms of the license and acknowledge that the information I provide will be collected stored processed and shared in accordance with the Meta Privacy Policy: checkbox
extra_gated_description: >-
The information you provide will be collected, stored, processed and shared in
accordance with the [Meta Privacy
Policy](https://www.facebook.com/privacy/policy/).
extra_gated_button_content: Submit
---
# <span style="color: #7FFF7F;">Llama-3.2-1B-Instruct GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.2-1B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.2-1B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.2-1B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.2-1B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.2-1B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.2-1B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.2-1B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.2-1B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.2-1B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.2-1B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.2-1B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
## Model Information
The Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks.
**Model Developer:** Meta
**Model Architecture:** Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
| | Training Data | Params | Input modalities | Output modalities | Context Length | GQA | Shared Embeddings | Token count | Knowledge cutoff |
| :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- | :---- |
| Llama 3.2 (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 128k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
| Llama 3.2 Quantized (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 8k | Yes | Yes | Up to 9T tokens | December 2023 |
| | | 3B (3.21B) | Multilingual Text | Multilingual Text and code | | | | | |
**Supported Languages:** English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages. Developers may fine-tune Llama 3.2 models for languages beyond these supported languages, provided they comply with the Llama 3.2 Community License and the Acceptable Use Policy. Developers are always expected to ensure that their deployments, including those that involve additional languages, are completed safely and responsibly.
**Llama 3.2 Model Family:** Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
**Model Release Date:** Sept 25, 2024
**Status:** This is a static model trained on an offline dataset. Future versions may be released that improve model capabilities and safety.
**License:** Use of Llama 3.2 is governed by the [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE) (a custom, commercial license agreement).
**Feedback:** Instructions on how to provide feedback or comments on the model can be found in the Llama Models [README](https://github.com/meta-llama/llama-models/blob/main/README.md). For more technical information about generation parameters and recipes for how to use Llama 3.2 in applications, please go [here](https://github.com/meta-llama/llama-recipes).
## Intended Use
**Intended Use Cases:** Llama 3.2 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat and agentic applications like knowledge retrieval and summarization, mobile AI powered writing assistants and query and prompt rewriting. Pretrained models can be adapted for a variety of additional natural language generation tasks. Similarly, quantized models can be adapted for a variety of on-device use-cases with limited compute resources.
**Out of Scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3.2 Community License. Use in languages beyond those explicitly referenced as supported in this model card.
## How to use
This repository contains two versions of Llama-3.2-1B-Instruct, for use with transformers and with the original `llama` codebase.
### Use with transformers
Starting with `transformers >= 4.43.0` onward, you can run conversational inference using the Transformers `pipeline` abstraction or by leveraging the Auto classes with the `generate()` function.
Make sure to update your transformers installation via `pip install --upgrade transformers`.
```python
import torch
from transformers import pipeline
model_id = "meta-llama/Llama-3.2-1B-Instruct"
pipe = pipeline(
"text-generation",
model=model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
outputs = pipe(
messages,
max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])
```
Note: You can also find detailed recipes on how to use the model locally, with `torch.compile()`, assisted generations, quantised and more at [`huggingface-llama-recipes`](https://github.com/huggingface/huggingface-llama-recipes)
### Use with `llama`
Please, follow the instructions in the [repository](https://github.com/meta-llama/llama)
To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
```
huggingface-cli download meta-llama/Llama-3.2-1B-Instruct --include "original/*" --local-dir Llama-3.2-1B-Instruct
```
## Hardware and Software
**Training Factors:** We used custom training libraries, Meta's custom built GPU cluster, and production infrastructure for pretraining. Fine-tuning, quantization, annotation, and evaluation were also performed on production infrastructure.
**Training Energy Use:** Training utilized a cumulative of **916k** GPU hours of computation on H100-80GB (TDP of 700W) type hardware, per the table below. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
**Training Greenhouse Gas Emissions:** Estimated total location-based greenhouse gas emissions were **240** tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with renewable energy; therefore, the total market-based greenhouse gas emissions for training were 0 tons CO2eq.
| | Training Time (GPU hours) | Logit Generation Time (GPU Hours) | Training Power Consumption (W) | Training Location-Based Greenhouse Gas Emissions (tons CO2eq) | Training Market-Based Greenhouse Gas Emissions (tons CO2eq) |
| :---- | :---: | ----- | :---: | :---: | :---: |
| Llama 3.2 1B | 370k | \- | 700 | 107 | 0 |
| Llama 3.2 3B | 460k | \- | 700 | 133 | 0 |
| Llama 3.2 1B SpinQuant | 1.7 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 3B SpinQuant | 2.4 | 0 | 700 | *Negligible*\*\* | 0 |
| Llama 3.2 1B QLora | 1.3k | 0 | 700 | 0.381 | 0 |
| Llama 3.2 3B QLora | 1.6k | 0 | 700 | 0.461 | 0 |
| Total | 833k | 86k | | 240 | 0 |
\*\* The location-based CO2e emissions of Llama 3.2 1B SpinQuant and Llama 3.2 3B SpinQuant are less than 0.001 metric tonnes each. This is due to the minimal training GPU hours that are required.
The methodology used to determine training energy use and greenhouse gas emissions can be found [here](https://arxiv.org/pdf/2204.05149). Since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others.
## Training Data
**Overview:** Llama 3.2 was pretrained on up to 9 trillion tokens of data from publicly available sources. For the 1B and 3B Llama 3.2 models, we incorporated logits from the Llama 3.1 8B and 70B models into the pretraining stage of the model development, where outputs (logits) from these larger models were used as token-level targets. Knowledge distillation was used after pruning to recover performance. In post-training we used a similar recipe as Llama 3.1 and produced final chat models by doing several rounds of alignment on top of the pre-trained model. Each round involved Supervised Fine-Tuning (SFT), Rejection Sampling (RS), and Direct Preference Optimization (DPO).
**Data Freshness:** The pretraining data has a cutoff of December 2023\.
## Quantization
### Quantization Scheme
We designed the current quantization scheme with the [PyTorch’s ExecuTorch](https://github.com/pytorch/executorch) inference framework and Arm CPU backend in mind, taking into account metrics including model quality, prefill/decoding speed, and memory footprint. Our quantization scheme involves three parts:
- All linear layers in all transformer blocks are quantized to a 4-bit groupwise scheme (with a group size of 32) for weights and 8-bit per-token dynamic quantization for activations.
- The classification layer is quantized to 8-bit per-channel for weight and 8-bit per token dynamic quantization for activation.
- Similar to classification layer, an 8-bit per channel quantization is used for embedding layer.
### Quantization-Aware Training and LoRA
The quantization-aware training (QAT) with low-rank adaptation (LoRA) models went through only post-training stages, using the same data as the full precision models. To initialize QAT, we utilize BF16 Llama 3.2 model checkpoints obtained after supervised fine-tuning (SFT) and perform an additional full round of SFT training with QAT. We then freeze the backbone of the QAT model and perform another round of SFT with LoRA adaptors applied to all layers within the transformer block. Meanwhile, the LoRA adaptors' weights and activations are maintained in BF16. Because our approach is similar to QLoRA of Dettmers et al., (2023) (i.e., quantization followed by LoRA adapters), we refer this method as QLoRA. Finally, we fine-tune the resulting model (both backbone and LoRA adaptors) using direct preference optimization (DPO).
### SpinQuant
[SpinQuant](https://arxiv.org/abs/2405.16406) was applied, together with generative post-training quantization (GPTQ). For the SpinQuant rotation matrix fine-tuning, we optimized for 100 iterations, using 800 samples with sequence-length 2048 from the WikiText 2 dataset. For GPTQ, we used 128 samples from the same dataset with the same sequence-length.
## Benchmarks \- English Text
In this section, we report the results for Llama 3.2 models on standard automatic benchmarks. For all these evaluations, we used our internal evaluations library.
### Base Pretrained Models
| Category | Benchmark | \# Shots | Metric | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
| ----- | ----- | :---: | :---: | :---: | :---: | :---: |
| General | MMLU | 5 | macro\_avg/acc\_char | 32.2 | 58 | 66.7 |
| | AGIEval English | 3-5 | average/acc\_char | 23.3 | 39.2 | 47.8 |
| | ARC-Challenge | 25 | acc\_char | 32.8 | 69.1 | 79.7 |
| Reading comprehension | SQuAD | 1 | em | 49.2 | 67.7 | 77 |
| | QuAC (F1) | 1 | f1 | 37.9 | 42.9 | 44.9 |
| | DROP (F1) | 3 | f1 | 28.0 | 45.2 | 59.5 |
| Long Context | Needle in Haystack | 0 | em | 96.8 | 1 | 1 |
### Instruction Tuned Models
| Capability | | Benchmark | \# Shots | Metric | Llama 3.2 1B bf16 | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B bf16 | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | ----- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | | MMLU | 5 | macro\_avg/acc | 49.3 | 43.3 | 47.3 | 49.0 | 63.4 | 60.5 | 62 | 62.4 | 69.4 |
| Re-writing | | Open-rewrite eval | 0 | micro\_avg/rougeL | 41.6 | 39.2 | 40.9 | 41.2 | 40.1 | 40.3 | 40.8 | 40.7 | 40.9 |
| Summarization | | TLDR9+ (test) | 1 | rougeL | 16.8 | 14.9 | 16.7 | 16.8 | 19.0 | 19.1 | 19.2 | 19.1 | 17.2 |
| Instruction following | | IFEval | 0 | Avg(Prompt/Instruction acc Loose/Strict) | 59.5 | 51.5 | 58.4 | 55.6 | 77.4 | 73.9 | 73.5 | 75.9 | 80.4 |
| Math | | GSM8K (CoT) | 8 | em\_maj1@1 | 44.4 | 33.1 | 40.6 | 46.5 | 77.7 | 72.9 | 75.7 | 77.9 | 84.5 |
| | | MATH (CoT) | 0 | final\_em | 30.6 | 20.5 | 25.3 | 31.0 | 48.0 | 44.2 | 45.3 | 49.2 | 51.9 |
| Reasoning | | ARC-C | 0 | acc | 59.4 | 54.3 | 57 | 60.7 | 78.6 | 75.6 | 77.6 | 77.6 | 83.4 |
| | | GPQA | 0 | acc | 27.2 | 25.9 | 26.3 | 25.9 | 32.8 | 32.8 | 31.7 | 33.9 | 32.8 |
| | | Hellaswag | 0 | acc | 41.2 | 38.1 | 41.3 | 41.5 | 69.8 | 66.3 | 68 | 66.3 | 78.7 |
| Tool Use | | BFCL V2 | 0 | acc | 25.7 | 14.3 | 15.9 | 23.7 | 67.0 | 53.4 | 60.1 | 63.5 | 67.1 |
| | | Nexus | 0 | macro\_avg/acc | 13.5 | 5.2 | 9.6 | 12.5 | 34.3 | 32.4 | 31.5 | 30.1 | 38.5 |
| Long Context | | InfiniteBench/En.QA | 0 | longbook\_qa/f1 | 20.3 | N/A | N/A | N/A | 19.8 | N/A | N/A | N/A | 27.3 |
| | | InfiniteBench/En.MC | 0 | longbook\_choice/acc | 38.0 | N/A | N/A | N/A | 63.3 | N/A | N/A | N/A | 72.2 |
| | | NIH/Multi-needle | 0 | recall | 75.0 | N/A | N/A | N/A | 84.7 | N/A | N/A | N/A | 98.8 |
| Multilingual | | MGSM (CoT) | 0 | em | 24.5 | 13.7 | 18.2 | 24.4 | 58.2 | 48.9 | 54.3 | 56.8 | 68.9 |
\*\*for comparison purposes only. Model not released.
### Multilingual Benchmarks
| Category | Benchmark | Language | Llama 3.2 1B | Llama 3.2 1B Vanilla PTQ\*\* | Llama 3.2 1B Spin Quant | Llama 3.2 1B QLoRA | Llama 3.2 3B | Llama 3.2 3B Vanilla PTQ\*\* | Llama 3.2 3B Spin Quant | Llama 3.2 3B QLoRA | Llama 3.1 8B |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| General | MMLU (5-shot, macro_avg/acc) | Portuguese | 39.8 | 34.9 | 38.9 | 40.2 | 54.5 | 50.9 | 53.3 | 53.4 | 62.1 |
| | | Spanish | 41.5 | 36.0 | 39.8 | 41.8 | 55.1 | 51.9 | 53.6 | 53.6 | 62.5 |
| | | Italian | 39.8 | 34.9 | 38.1 | 40.6 | 53.8 | 49.9 | 52.1 | 51.7 | 61.6 |
| | | German | 39.2 | 34.9 | 37.5 | 39.6 | 53.3 | 50.0 | 52.2 | 51.3 | 60.6 |
| | | French | 40.5 | 34.8 | 39.2 | 40.8 | 54.6 | 51.2 | 53.3 | 53.3 | 62.3 |
| | | Hindi | 33.5 | 30.0 | 32.1 | 34.0 | 43.3 | 40.4 | 42.0 | 42.1 | 50.9 |
| | | Thai | 34.7 | 31.2 | 32.4 | 34.9 | 44.5 | 41.3 | 44.0 | 42.2 | 50.3 |
\*\*for comparison purposes only. Model not released.
## Inference time
In the below table, we compare the performance metrics of different quantization methods (SpinQuant and QAT \+ LoRA) with the BF16 baseline. The evaluation was done using the [ExecuTorch](https://github.com/pytorch/executorch) framework as the inference engine, with the ARM CPU as a backend using Android OnePlus 12 device.
| Category | Decode (tokens/sec) | Time-to-first-token (sec) | Prefill (tokens/sec) | Model size (PTE file size in MB) | Memory size (RSS in MB) |
| :---- | ----- | ----- | ----- | ----- | ----- |
| 1B BF16 (baseline) | 19.2 | 1.0 | 60.3 | 2358 | 3,185 |
| 1B SpinQuant | 50.2 (2.6x) | 0.3 (-76.9%) | 260.5 (4.3x) | 1083 (-54.1%) | 1,921 (-39.7%) |
| 1B QLoRA | 45.8 (2.4x) | 0.3 (-76.0%) | 252.0 (4.2x) | 1127 (-52.2%) | 2,255 (-29.2%) |
| 3B BF16 (baseline) | 7.6 | 3.0 | 21.2 | 6129 | 7,419 |
| 3B SpinQuant | 19.7 (2.6x) | 0.7 (-76.4%) | 89.7 (4.2x) | 2435 (-60.3%) | 3,726 (-49.8%) |
| 3B QLoRA | 18.5 (2.4x) | 0.7 (-76.1%) | 88.8 (4.2x) | 2529 (-58.7%) | 4,060 (-45.3%) |
(\*) The performance measurement is done using an adb binary-based approach.
(\*\*) It is measured on an Android OnePlus 12 device.
(\*\*\*) Time-to-first-token (TTFT) is measured with prompt length=64
*Footnote:*
- *Decode (tokens/second) is for how quickly it keeps generating. Higher is better.*
- *Time-to-first-token (TTFT for shorthand) is for how fast it generates the first token for a given prompt. Lower is better.*
- *Prefill is the inverse of TTFT (aka 1/TTFT) in tokens/second. Higher is better*
- *Model size \- how big is the model, measured by, PTE file, a binary file format for ExecuTorch*
- *RSS size \- Memory usage in resident set size (RSS)*
## Responsibility & Safety
As part of our Responsible release approach, we followed a three-pronged strategy to managing trust & safety risks:
1. Enable developers to deploy helpful, safe and flexible experiences for their target audience and for the use cases supported by Llama
2. Protect developers against adversarial users aiming to exploit Llama capabilities to potentially cause harm
3. Provide protections for the community to help prevent the misuse of our models
### Responsible Deployment
**Approach:** Llama is a foundational technology designed to be used in a variety of use cases. Examples on how Meta’s Llama models have been responsibly deployed can be found in our [Community Stories webpage](https://llama.meta.com/community-stories/). Our approach is to build the most helpful models, enabling the world to benefit from the technology power, by aligning our model safety for generic use cases and addressing a standard set of harms. Developers are then in the driver’s seat to tailor safety for their use cases, defining their own policies and deploying the models with the necessary safeguards in their Llama systems. Llama 3.2 was developed following the best practices outlined in our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide/).
#### Llama 3.2 Instruct
**Objective:** Our main objectives for conducting safety fine-tuning are to provide the research community with a valuable resource for studying the robustness of safety fine-tuning, as well as to offer developers a readily available, safe, and powerful model for various applications to reduce the developer workload to deploy safe AI systems. We implemented the same set of safety mitigations as in Llama 3, and you can learn more about these in the Llama 3 [paper](https://ai.meta.com/research/publications/the-llama-3-herd-of-models/).
**Fine-Tuning Data:** We employ a multi-faceted approach to data collection, combining human-generated data from our vendors with synthetic data to mitigate potential safety risks. We’ve developed many large language model (LLM)-based classifiers that enable us to thoughtfully select high-quality prompts and responses, enhancing data quality control.
**Refusals and Tone:** Building on the work we started with Llama 3, we put a great emphasis on model refusals to benign prompts as well as refusal tone. We included both borderline and adversarial prompts in our safety data strategy, and modified our safety data responses to follow tone guidelines.
#### Llama 3.2 Systems
**Safety as a System:** Large language models, including Llama 3.2, **are not designed to be deployed in isolation** but instead should be deployed as part of an overall AI system with additional safety guardrails as required. Developers are expected to deploy system safeguards when building agentic systems. Safeguards are key to achieve the right helpfulness-safety alignment as well as mitigating safety and security risks inherent to the system and any integration of the model or system with external tools. As part of our responsible release approach, we provide the community with [safeguards](https://llama.meta.com/trust-and-safety/) that developers should deploy with Llama models or other LLMs, including Llama Guard, Prompt Guard and Code Shield. All our [reference implementations](https://github.com/meta-llama/llama-agentic-system) demos contain these safeguards by default so developers can benefit from system-level safety out-of-the-box.
### New Capabilities and Use Cases
**Technological Advancement:** Llama releases usually introduce new capabilities that require specific considerations in addition to the best practices that generally apply across all Generative AI use cases. For prior release capabilities also supported by Llama 3.2, see [Llama 3.1 Model Card](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md), as the same considerations apply here as well.
**Constrained Environments:** Llama 3.2 1B and 3B models are expected to be deployed in highly constrained environments, such as mobile devices. LLM Systems using smaller models will have a different alignment profile and safety/helpfulness tradeoff than more complex, larger systems. Developers should ensure the safety of their system meets the requirements of their use case. We recommend using lighter system safeguards for such use cases, like Llama Guard 3-1B or its mobile-optimized version.
### Evaluations
**Scaled Evaluations:** We built dedicated, adversarial evaluation datasets and evaluated systems composed of Llama models and Purple Llama safeguards to filter input prompt and output response. It is important to evaluate applications in context, and we recommend building dedicated evaluation dataset for your use case.
**Red Teaming:** We conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we used the learnings to improve our benchmarks and safety tuning datasets. We partnered early with subject-matter experts in critical risk areas to understand the nature of these real-world harms and how such models may lead to unintended harm for society. Based on these conversations, we derived a set of adversarial goals for the red team to attempt to achieve, such as extracting harmful information or reprogramming the model to act in a potentially harmful capacity. The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity in addition to multilingual content specialists with background in integrity issues in specific geographic markets.
### Critical Risks
In addition to our safety work above, we took extra care on measuring and/or mitigating the following critical risk areas:
**1\. CBRNE (Chemical, Biological, Radiological, Nuclear, and Explosive Weapons):** Llama 3.2 1B and 3B models are smaller and less capable derivatives of Llama 3.1. For Llama 3.1 70B and 405B, to assess risks related to proliferation of chemical and biological weapons, we performed uplift testing designed to assess whether use of Llama 3.1 models could meaningfully increase the capabilities of malicious actors to plan or carry out attacks using these types of weapons and have determined that such testing also applies to the smaller 1B and 3B models.
**2\. Child Safety:** Child Safety risk assessments were conducted using a team of experts, to assess the model’s capability to produce outputs that could result in Child Safety risks and inform on any necessary and appropriate risk mitigations via fine tuning. We leveraged those expert red teaming sessions to expand the coverage of our evaluation benchmarks through Llama 3 model development. For Llama 3, we conducted new in-depth sessions using objective based methodologies to assess the model risks along multiple attack vectors including the additional languages Llama 3 is trained on. We also partnered with content specialists to perform red teaming exercises assessing potentially violating content while taking account of market specific nuances or experiences.
**3\. Cyber Attacks:** For Llama 3.1 405B, our cyber attack uplift study investigated whether LLMs can enhance human capabilities in hacking tasks, both in terms of skill level and speed.
Our attack automation study focused on evaluating the capabilities of LLMs when used as autonomous agents in cyber offensive operations, specifically in the context of ransomware attacks. This evaluation was distinct from previous studies that considered LLMs as interactive assistants. The primary objective was to assess whether these models could effectively function as independent agents in executing complex cyber-attacks without human intervention. Because Llama 3.2’s 1B and 3B models are smaller and less capable models than Llama 3.1 405B, we broadly believe that the testing conducted for the 405B model also applies to Llama 3.2 models.
### Community
**Industry Partnerships:** Generative AI safety requires expertise and tooling, and we believe in the strength of the open community to accelerate its progress. We are active members of open consortiums, including the AI Alliance, Partnership on AI and MLCommons, actively contributing to safety standardization and transparency. We encourage the community to adopt taxonomies like the MLCommons Proof of Concept evaluation to facilitate collaboration and transparency on safety and content evaluations. Our Purple Llama tools are open sourced for the community to use and widely distributed across ecosystem partners including cloud service providers. We encourage community contributions to our [Github repository](https://github.com/meta-llama/PurpleLlama).
**Grants:** We also set up the [Llama Impact Grants](https://llama.meta.com/llama-impact-grants/) program to identify and support the most compelling applications of Meta’s Llama model for societal benefit across three categories: education, climate and open innovation. The 20 finalists from the hundreds of applications can be found [here](https://llama.meta.com/llama-impact-grants/#finalists).
**Reporting:** Finally, we put in place a set of resources including an [output reporting mechanism](https://developers.facebook.com/llama_output_feedback) and [bug bounty program](https://www.facebook.com/whitehat) to continuously improve the Llama technology with the help of the community.
## Ethical Considerations and Limitations
**Values:** The core values of Llama 3.2 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3.2 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
**Testing:** Llama 3.2 is a new technology, and like any new technology, there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all scenarios. For these reasons, as with all LLMs, Llama 3.2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 3.2 models, developers should perform safety testing and tuning tailored to their specific applications of the model. Please refer to available resources including our [Responsible Use Guide](https://llama.meta.com/responsible-use-guide), [Trust and Safety](https://llama.meta.com/trust-and-safety/) solutions, and other [resources](https://llama.meta.com/docs/get-started/) to learn more about responsible development.
|
Mungert/Phi-4-mini-reasoning-GGUF | Mungert | "2025-05-06T22:58:00Z" | 1,112 | 1 | transformers | [
"transformers",
"gguf",
"nlp",
"math",
"code",
"text-generation",
"en",
"arxiv:2504.21233",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-05-02T18:21:44Z" | ---
language:
- en
library_name: transformers
license: mit
license_link: https://huggingface.co/microsoft/Phi-4-mini-instruct-reasoning/resolve/main/LICENSE
pipeline_tag: text-generation
tags:
- nlp
- math
- code
widget:
- messages:
- role: user
content: How to solve 3*x^2+4*x+5=1?
---
# <span style="color: #7FFF7F;">Phi-4-mini-reasoning GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Phi-4-mini-reasoning-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Phi-4-mini-reasoning-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Phi-4-mini-reasoning-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Phi-4-mini-reasoning-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Phi-4-mini-reasoning-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Phi-4-mini-reasoning-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Phi-4-mini-reasoning-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Phi-4-mini-reasoning-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Phi-4-mini-reasoning-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Phi-4-mini-reasoning-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Phi-4-mini-reasoning-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard/?assistant=open)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Free Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by logging in or [downloading our Free Network Monitor Agent with integrated AI Assistant](https://readyforquantum.com/download)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Free Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
## Model Summary
Phi-4-mini-reasoning is a lightweight open model built upon synthetic data with a focus on high-quality, reasoning dense data further finetuned for more advanced math reasoning capabilities.
The model belongs to the Phi-4 model family and supports 128K token context length.
📰 [Phi-4-mini-reasoning Blog](https://aka.ms/phi4-mini-reasoning/blog), and [Developer Article](https://techcommunity.microsoft.com/blog/azuredevcommunityblog/make-phi-4-mini-reasoning-more-powerful-with-industry-reasoning-on-edge-devices/4409764)<br>
📖 [Phi-4-mini-reasoning Technical Report](https://aka.ms/phi4-mini-reasoning/techreport) | [HF paper](https://huggingface.co/papers/2504.21233) <br>
👩🍳 [Phi Cookbook](https://github.com/microsoft/PhiCookBook) <br>
🏡 [Phi Portal](https://azure.microsoft.com/en-us/products/phi) <br>
🖥️ Try It [Azure](https://aka.ms/phi4-mini-reasoning/azure) <br>
🎉**Phi-4 models**: [[Phi-4-reasoning](https://huggingface.co/microsoft/Phi-4-reasoning)] | [[multimodal-instruct](https://huggingface.co/microsoft/Phi-4-multimodal-instruct) | [onnx](https://huggingface.co/microsoft/Phi-4-multimodal-instruct-onnx)];
[[mini-instruct](https://huggingface.co/microsoft/Phi-4-mini-instruct) | [onnx](https://huggingface.co/microsoft/Phi-4-mini-instruct-onnx)]
## Intended Uses
### Primary Use Cases
Phi-4-mini-reasoning is designed for multi-step, logic-intensive mathematical problem-solving tasks under memory/compute constrained environments and latency bound scenarios.
Some of the use cases include formal proof generation, symbolic computation, advanced word problems, and a wide range of mathematical reasoning scenarios.
These models excel at maintaining context across steps, applying structured logic, and delivering accurate, reliable solutions in domains that require deep analytical thinking.
### Use Case Considerations
This model is designed and tested for math reasoning only. It is not specifically designed or evaluated for all downstream purposes.
Developers should consider common limitations of language models, as well as performance difference across languages, as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios.
Developers should be aware of and adhere to applicable laws or regulations (including but not limited to privacy, trade compliance laws, etc.) that are relevant to their use case.
***Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.***
## Release Notes
This release of Phi-4-mini-reasoning addresses user feedback and market demand for a compact reasoning model.
It is a compact transformer-based language model optimized for mathematical reasoning, built to deliver high-quality, step-by-step problem solving in environments where computing or latency is constrained.
The model is fine-tuned with synthetic math data from a more capable model (much larger, smarter, more accurate, and better at following instructions), which has resulted in enhanced reasoning performance.
Phi-4-mini-reasoning balances reasoning ability with efficiency, making it potentially suitable for educational applications, embedded tutoring, and lightweight deployment on edge or mobile systems.
If a critical issue is identified with Phi-4-mini-reasoning, it should be promptly reported through the MSRC Researcher Portal or [email protected]
### Model Quality
To understand the capabilities, the 3.8B parameters Phi-4-mini-reasoning model was compared with a set of models over a variety of reasoning benchmarks.
A high-level overview of the model quality is as follows:
| Model | AIME | MATH-500 | GPQA Diamond |
|------------------------------------|-------|----------|--------------|
| o1-mini* | 63.6 | 90.0 | 60.0 |
| DeepSeek-R1-Distill-Qwen-7B | 53.3 | 91.4 | 49.5 |
| DeepSeek-R1-Distill-Llama-8B | 43.3 | 86.9 | 47.3 |
| Bespoke-Stratos-7B* | 20.0 | 82.0 | 37.8 |
| OpenThinker-7B* | 31.3 | 83.0 | 42.4 |
| Llama-3.2-3B-Instruct | 6.7 | 44.4 | 25.3 |
| Phi-4-Mini (base model, 3.8B) | 10.0 | 71.8 | 36.9 |
|**Phi-4-mini-reasoning (3.8B)** | **57.5** | **94.6** | **52.0** |
Overall, the model with only 3.8B-param achieves a similar level of multilingual language understanding and reasoning ability as much larger models.
However, it is still fundamentally limited by its size for certain tasks. The model simply does not have the capacity to store too much factual knowledge, therefore, users may experience factual incorrectness. However, it may be possible to resolve such weakness by augmenting Phi-4 with a search engine, particularly when using the model under RAG settings.
## Usage
### Tokenizer
Phi-4-mini-reasoning supports a vocabulary size of up to `200064` tokens. The [tokenizer files](https://huggingface.co/microsoft/Phi-4-mini-reasoning/blob/main/added_tokens.json) already provide placeholder tokens that can be used for downstream fine-tuning, but they can also be extended up to the model's vocabulary size.
### Input Formats
Given the nature of the training data, the Phi-4-mini-instruct
model is best suited for prompts using specific formats.
Below are the two primary formats:
#### Chat format
This format is used for general conversation and instructions:
```yaml
<|system|>Your name is Phi, an AI math expert developed by Microsoft.<|end|><|user|>How to solve 3*x^2+4*x+5=1?<|end|><|assistant|>
```
### Inference with transformers
Phi-4-mini-reasoning has been integrated in the `4.51.3` version of `transformers`. The current `transformers` version can be verified with: `pip list | grep transformers`.
Python 3.8 and 3.10 will work best.
List of required packages:
```
flash_attn==2.7.4.post1
torch==2.5.1
transformers==4.51.3
accelerate==1.3.0
```
Phi-4-mini-reasoning is also available in [Azure AI Studio](https://aka.ms/phi-4-mini-reasoning/azure)
#### Example
After obtaining the Phi-4-mini-instruct model checkpoints, users can use this sample code for inference.
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
torch.random.manual_seed(0)
model_id = "microsoft/Phi-4-mini-reasoning"
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map="cuda",
torch_dtype="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
messages = [{
"role": "user",
"content": "How to solve 3*x^2+4*x+5=1?"
}]
inputs = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
)
outputs = model.generate(
**inputs.to(model.device),
max_new_tokens=32768,
temperature=0.8,
top_p=0.95,
do_sample=True,
)
outputs = tokenizer.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])
print(outputs[0])
```
## Training
### Model
+ **Architecture:** Phi-4-mini-reasoning shares the same architecture as Phi-4-Mini, which has 3.8B parameters and is a dense decoder-only Transformer model. When compared with Phi-3.5-Mini, the major changes with Phi-4-Mini are 200K vocabulary, grouped-query attention, and shared input and output embedding.<br>
+ **Inputs:** Text. It is best suited for prompts using the chat format.<br>
+ **Context length:** 128K tokens<br>
+ **GPUs:** 128 H100-80G<br>
+ **Training time:** 2 days<br>
+ **Training data:** 150B tokens<br>
+ **Outputs:** Generated text<br>
+ **Dates:** Trained in February 2024<br>
+ **Status:** This is a static model trained on offline datasets with the cutoff date of February 2025 for publicly available data.<br>
+ **Supported languages:** English<br>
+ **Release date:** April 2025<br>
### Training Datasets
The training data for Phi-4-mini-reasoning consists exclusively of synthetic mathematical content generated by a stronger and more advanced reasoning model, Deepseek-R1.
The objective is to distill knowledge from this model. This synthetic dataset comprises over one million diverse math problems spanning multiple levels of difficulty (from middle school to Ph.D. level).
For each problem in the synthetic dataset, eight distinct solutions (rollouts) were sampled, and only those verified as correct were retained, resulting in approximately 30 billion tokens of math content.
The dataset integrates three primary components:
1) a curated selection of high-quality, publicly available math questions and a part of the SFT(Supervised Fine-Tuning) data that was used to train the base Phi-4-Mini model;
2) an extensive collection of synthetic math data generated by the Deepseek-R1 model, designed specifically for high-quality supervised fine-tuning and model distillation; and
3) a balanced set of correct and incorrect answers used to construct preference data aimed at enhancing Phi-4-mini-reasoning's reasoning capabilities by learning more effective reasoning trajectories
## Software
* [PyTorch](https://github.com/pytorch/pytorch)
* [Transformers](https://github.com/huggingface/transformers)
* [Flash-Attention](https://github.com/HazyResearch/flash-attention)
## Hardware
Note that by default, the Phi-4-mini-reasoning model uses flash attention, which requires certain types of GPU hardware to run. We have tested on the following GPU types:
* NVIDIA A100
* NVIDIA H100
If you want to run the model on:
* NVIDIA V100 or earlier generation GPUs: call AutoModelForCausalLM.from_pretrained() with attn_implementation="eager"
## Safety Evaluation and Red-Teaming
The Phi-4 family of models has adopted a robust safety post-training approach. This approach leverages a variety of both open-source and in-house generated datasets. The overall technique employed to do the safety alignment is a combination of SFT, DPO (Direct Preference Optimization), and RLHF (Reinforcement Learning from Human Feedback) approaches by utilizing human-labeled and synthetic English-language datasets, including publicly available datasets focusing on helpfulness and harmlessness, as well as various questions and answers targeted to multiple safety categories.
Phi-4-Mini-Reasoning was developed in accordance with Microsoft's responsible AI principles. Potential safety risks in the model’s responses were assessed using the Azure AI Foundry’s Risk and Safety Evaluation framework, focusing on harmful content, direct jailbreak, and model groundedness. The Phi-4-Mini-Reasoning Model Card contains additional information about our approach to safety and responsible AI considerations that developers should be aware of when using this model.
## Responsible AI Considerations
Like other language models, the Phi family of models can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include:
+ Quality of Service: The Phi models are trained primarily on English text and some additional multilingual text. Languages other than English will experience worse performance as well as performance disparities across non-English. English language varieties with less representation in the training data might experience worse performance than standard American English.
+ Multilingual performance and safety gaps: We believe it is important to make language models more widely available across different languages, but the Phi 4 models still exhibit challenges common across multilingual releases. As with any deployment of LLMs, developers will be better positioned to test for performance or safety gaps for their linguistic and cultural context and customize the model with additional fine-tuning and appropriate safeguards.
+ Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups, cultural contexts, or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases.
+ Inappropriate or Offensive Content: These models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the case.
+ Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated.
+ Election Information Reliability : The model has an elevated defect rate when responding to election-critical queries, which may result in incorrect or unauthoritative election critical information being presented. We are working to improve the model's performance in this area. Users should verify information related to elections with the election authority in their region.
+ Limited Scope for Code: The majority of Phi 4 training data is based in Python and uses common packages such as "typing, math, random, collections, datetime, itertools". If the model generates Python scripts that utilize other packages or scripts in other languages, it is strongly recommended that users manually verify all API uses.
+ Long Conversation: Phi 4 models, like other models, can in some cases generate responses that are repetitive, unhelpful, or inconsistent in very long chat sessions in both English and non-English languages. Developers are encouraged to place appropriate mitigations, like limiting conversation turns to account for the possible conversational drift.
Developers should apply responsible AI best practices, including mapping, measuring, and mitigating risks associated with their specific use case and cultural, linguistic context. Phi 4 family of models are general purpose models. As developers plan to deploy these models for specific use cases, they are encouraged to fine-tune the models for their use case and leverage the models as part of broader AI systems with language-specific safeguards in place. Important areas for consideration include:
+ Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques.
+ High-Risk Scenarios: Developers should assess the suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context.
+ Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).
+ Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
+ Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.
## License
The model is licensed under the [MIT license](./LICENSE).
## Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow [Microsoft’s Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks). Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.
## Appendix A: Benchmark Methodology
We include a brief word on methodology here - and in particular, how we think about optimizing prompts. In an ideal world, we would never change any prompts in our benchmarks to ensure it is always an apples-to-apples comparison when comparing different models. Indeed, this is our default approach, and is the case in the vast majority of models we have run to date. For all benchmarks, we consider using the same generation configuration such as max sequence length (32768), the same temperature for the fair comparison.
Benchmark datasets
We evaluate the model with three of the most popular math benchmarks where the strongest reasoning models are competing together. Specifically:
- Math-500: This benchmark consists of 500 challenging math problems designed to test the model's ability to perform complex mathematical reasoning and problem-solving.
- AIME 2024: The American Invitational Mathematics Examination (AIME) is a highly regarded math competition that features a series of difficult problems aimed at assessing advanced mathematical skills and logical reasoning.
- GPQA Diamond: The Graduate-Level Google-Proof Q&A (GPQA) Diamond benchmark focuses on evaluating the model's ability to understand and solve a wide range of mathematical questions, including both straightforward calculations and more intricate problem-solving tasks.
|
Mungert/Foundation-Sec-8B-GGUF | Mungert | "2025-05-06T22:57:57Z" | 2,199 | 1 | transformers | [
"transformers",
"gguf",
"security",
"text-generation",
"en",
"arxiv:2504.21039",
"base_model:meta-llama/Llama-3.1-8B",
"base_model:quantized:meta-llama/Llama-3.1-8B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix"
] | text-generation | "2025-05-01T17:36:08Z" | ---
license: apache-2.0
language:
- en
base_model:
- meta-llama/Llama-3.1-8B
pipeline_tag: text-generation
library_name: transformers
tags:
- security
---
# <span style="color: #7FFF7F;">Foundation-Sec-8B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Foundation-Sec-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Foundation-Sec-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Foundation-Sec-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Foundation-Sec-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Foundation-Sec-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Foundation-Sec-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Foundation-Sec-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Foundation-Sec-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Foundation-Sec-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Foundation-Sec-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Foundation-Sec-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard/?assistant=open)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Free Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by logging in or [downloading our Free Network Monitor Agent with integrated AI Assistant](https://readyforquantum.com/download)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Free Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
# Foundation-Sec-8B - Model Card
## Model Information
Foundation-Sec-8B (Llama-3.1-FoundationAI-SecurityLLM-base-8B) is an open-weight, 8-billion parameter base language model specialized for cybersecurity applications. It extends Llama-3.1-8B model through continued pretraining on a curated corpus of cybersecurity-specific text, including threat intelligence reports, vulnerability databases, incident response documentation, and security standards. It has been trained to understand security concepts, terminology, and practices across multiple security domains. The model is designed to serve as a domain-adapted base model for use in applications such as threat detection, vulnerability assessment, security automation, and attack simulation. Foundation-Sec-8B enables organizations to build AI-driven security tools that can be deployed locally, reducing dependency on cloud-based AI services while maintaining high performance on security-related tasks.
- **Model Name:** Foundation-Sec-8B (Llama-3.1-FoundationAI-SecurityLLM-base-8B)
- **Model Developer:** Amin Karbasi and team at Foundation AI — Cisco
- **Technical Report:** [`https://arxiv.org/abs/2504.21039`](https://arxiv.org/abs/2504.21039)
- **Model Card Contact:** For questions about the team, model usage, and future directions, contact [`[email protected]`](mailto:[email protected]). For technical questions about the model, please contact [`[email protected]`](mailto:[email protected]).
- **Model Release Date:** April 28, 2025
- **Supported Language(s):** English
- **Model Architecture:** Auto-regressive language model that uses an optimized transformer architecture (Meta Llama-3.1-8B backbone)
- **Training Objective:** Continued pre-training on cybersecurity-specific corpus
- **Training Data Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released on updated data.
- **License:** Apache 2.0
## Intended Use
### Intended Use Cases
Foundation-Sec-8B is designed for security practitioners, researchers, and developers building AI-powered security workflows and applications. Foundation-Sec-8B is optimized for three core use case categories:
- **SOC Acceleration**: Automating triage, summarization, case note generation, and evidence collection.
- **Proactive Threat Defense**: Simulating attacks, prioritizing vulnerabilities, mapping TTPs, and modeling attacker behavior.
- **Engineering Enablement**: Providing security assistance, validating configurations, assessing compliance evidence, and improving security posture.
The model is intended for local deployment in environments prioritizing data security, regulatory compliance, and operational control.
### Downstream Use
Foundation-Sec-8B can be used directly for security-related language tasks and serves as a strong starting point for fine-tuning across a variety of cybersecurity workflows. Example downstream applications include:
- Summarization
- Summarizing detection playbooks and incident reports
- Consolidating fragmented analyst notes into structured case summaries
- Classification
- Mapping threats to MITRE ATT&CK techniques
- Prioritizing vulnerabilities based on contextual risk
- Classifying security-relevant emails and leaked file contents
- Named Entity Recognition
- Extracting compliance evidence from documents
- Building network behavior profiles from technical manuals
- Question & Answer
- Assisting SOC analysts with alert triage and investigation
- Responding to cloud security and software compliance queries
- Reasoning and Text Generation
- Generating red-team attack plans and threat models
- Predicting attacker next steps in active investigations
- Enriching vulnerability scan results with contextual insights
For questions or assistance with fine-tuning Foundation-Sec-8B, please contact **Paul Kassianik** ([email protected]) or **Dhruv Kedia** ([email protected]).
### Out-of-Scope Use
The following uses are out-of-scope and are neither recommended nor intended use cases:
1. **Generating harmful content** - The model should not be used to:
- Generate malware or other malicious code
- Create phishing content or social engineering scripts
- Develop attack plans targeting specific organizations
- Design exploitation techniques for vulnerabilities without legitimate security research purposes
2. **Critical security decisions without human oversight** - The model should not be used for:
- Autonomous security decision-making without human review
- Critical infrastructure protection without expert supervision
- Final determination of security compliance without human verification
- Autonomous vulnerability remediation without testing
3. **Legal or medical advice** - The model is not qualified to provide:
- Legal advice regarding security regulations, compliance requirements, or intellectual property disputes
- Legal advice regarding security issues that would reference legal statutes, precedents, or case law necessary to provide legal advice
- Medical advice regarding health impacts of security incidents
4. **Non-security use cases** - The model is specifically optimized for cybersecurity and may not perform as well on general tasks as models trained for broader applications.
5. **Violation of Laws or Regulations** - Any use that violates applicable laws or regulations.
## How to Get Started with the Model
Use the code below to get started with the model.
```python
# Import the required libraries
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("fdtn-ai/Foundation-Sec-8B")
model = AutoModelForCausalLM.from_pretrained("fdtn-ai/Foundation-Sec-8B")
# Example: Matching CWE to CVE IDs
prompt="""CVE-2021-44228 is a remote code execution flaw in Apache Log4j2 via unsafe JNDI lookups (“Log4Shell”). The CWE is CWE-502.
CVE-2017-0144 is a remote code execution vulnerability in Microsoft’s SMBv1 server (“EternalBlue”) due to a buffer overflow. The CWE is CWE-119.
CVE-2014-0160 is an information-disclosure bug in OpenSSL’s heartbeat extension (“Heartbleed”) causing out-of-bounds reads. The CWE is CWE-125.
CVE-2017-5638 is a remote code execution issue in Apache Struts 2’s Jakarta Multipart parser stemming from improper input validation of the Content-Type header. The CWE is CWE-20.
CVE-2019-0708 is a remote code execution vulnerability in Microsoft’s Remote Desktop Services (“BlueKeep”) triggered by a use-after-free. The CWE is CWE-416.
CVE-2015-10011 is a vulnerability about OpenDNS OpenResolve improper log output neutralization. The CWE is"""
# Tokenize the input
inputs = tokenizer(prompt, return_tensors="pt")
# Generate the response
outputs = model.generate(
inputs["input_ids"],
max_new_tokens=3,
do_sample=True,
temperature=0.1,
top_p=0.9,
)
# Decode and print the response
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
response = response.replace(prompt, "").strip()
print(response)
```
## Training and Evaluation
### Training Data
Foundation-sec-8B was pretrained on approximately **5.1 billion tokens** of cybersecurity-specific data curated in-house by Cisco’s Foundation AI team. The dataset was meticulously collected from public sources on the web.
The pre-training corpus was built through a multi-stage pipeline that included large-scale web crawling, relevancy filtering, deduplication, and quality filtering.
**Data cutoff:** April 10th, 2025.
More detailed methodology is available in the technical report.
### Training Setup
Foundation-sec-8B is based on the **Llama 3.1 8B** architecture. Pre-training was performed on Cisco Foundation AI’s internal compute cluster.
Key training details:
- **Continued pretraining** for cybersecurity specialization
- **4096-token** sequence length
- **Optimizer:** AdamW
More detailed methodology is available in the technical report.
### Evaluation
Foundation-sec-8B was benchmarked on cybersecurity and general reasoning tasks, using a standardized 5-shot prompting setup (temperature = 0.3).
| **Benchmark** | **Foundation-sec-8B** | **Llama 3.1 8B** | **Llama 3.1 70B** |
| --- | --- | --- | --- |
| CTI-MCQA | 67.39 | 64.14 | 68.23 |
| CTI-RCM | 75.26 | 66.43 | 72.66 |
**Benchmark Overview:**
- **CTI-MCQA:** 2,500 multiple-choice questions testing cybersecurity knowledge across frameworks like MITRE ATT&CK, NIST, GDPR, and threat intelligence best practices.
- **CTI-RCM:** 900+ vulnerability root cause mapping examples linking CVEs to CWE categories, assessing deep understanding of security weaknesses.
**Key highlights:**
- **+3 to +9 point gains** over Llama-3.1-8B across security-specific benchmarks.
- **Comparable or better** performance than Llama-3.1-70B on cyber threat intelligence tasks.
- **Minimal drop (~2%)** in general language reasoning (MMLU) despite cybersecurity specialization.
For full benchmark details and evaluation methodology, please refer to the technical report.
## Limitations
Foundation-Sec-8B has several limitations that users should be aware of:
1. **Domain-specific knowledge limitations**:
- Foundation-Sec-8B may not be familiar with recent vulnerabilities, exploits, or novel attack vectors or security technologies released after its training cutoff date
- Knowledge of specialized or proprietary security systems or tools may be limited
2. **Potential biases**:
- The model may reflect biases present in security literature and documentation
- The model may be trained on known attack patterns and have difficulty recognizing novel attack vectors
- Security practices and recommendations may be biased toward certain technological ecosystems
- Geographic and cultural biases in security approaches may be present
3. **Security risks**:
- The model cannot verify the identity or intentions of users
- Adversarial prompting techniques might potentially bypass safety mechanisms
- The model may unintentionally provide information that could be misused if proper prompting guardrails are not implemented
4. **Contextual blindness:**
- The model may struggle to understand the complex interrelationships between systems, users, and data in order to provide accurate context.
5. **Technical limitations**:
- Performance varies based on how security concepts are described in prompts
- May not fully understand complex, multi-step security scenarios without clear explanation
- Cannot access external systems or actively scan environments
- Cannot independently verify factual accuracy of its outputs
6. **Ethical considerations**:
- Dual-use nature of security knowledge requires careful consideration of appropriate use cases
### Recommendations
To address the limitations of Foundation-Sec-8B, we recommend:
1. **Human oversight**:
- Always have qualified security professionals review model outputs before implementation
- Use the model as an assistive tool rather than a replacement for expert human judgment
- Implement a human-in-the-loop approach for security-critical applications
2. **System design safeguards**:
- Implement additional validation layers for applications built with this model
- Consider architectural constraints that limit the model's ability to perform potentially harmful actions (excessive agency)
- Deploy the model in environments with appropriate access controls
3. **Prompt engineering**:
- Use carefully designed prompts that encourage ethical security practices
- Include explicit instructions regarding responsible disclosure and ethical hacking principles
- Structure interactions to minimize the risk of inadvertently harmful outputs
4. **Knowledge supplementation**:
- Supplement the model with up-to-date security feeds and databases
- Implement retrieval-augmented generation for current threat intelligence sources
5. **Usage policies**:
- Develop and enforce clear acceptable use policies for applications using this model
- Implement monitoring and auditing for high-risk applications
- Create documentation for end users about the model's limitations |
Mungert/Qwen3-32B-GGUF | Mungert | "2025-05-06T22:57:53Z" | 1,075 | 1 | transformers | [
"transformers",
"gguf",
"text-generation",
"arxiv:2309.00071",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-05-01T13:56:25Z" | ---
library_name: transformers
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen3-32B/blob/main/LICENSE
pipeline_tag: text-generation
---
# <span style="color: #7FFF7F;">Qwen3-32B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Qwen3-32B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Qwen3-32B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Qwen3-32B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Qwen3-32B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Qwen3-32B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Qwen3-32B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Qwen3-32B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Qwen3-32B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Qwen3-32B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Qwen3-32B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Qwen3-32B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard/?assistant=open)
💬 **How to test**:
Choose an **AI assistant type**:
- `TurboLLM` (GPT-4o-mini)
- `HugLLM` (Hugginface Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Network Monitoring tasks**
🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4o-mini** for:
- **Create custom cmd processors to run .net code on Free Network Monitor Agents**
- **Real-time network diagnostics and monitoring**
- **Security Audits**
- **Penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by logging in or [downloading our Free Network Monitor Agent with integrated AI Assistant](https://readyforquantum.com/download)
🔵 **HugLLM** – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example commands to you could test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a comprehensive security audit on my server"`
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Free Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
# Qwen3-32B
<a href="https://chat.qwen.ai/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
</a>
## Qwen3 Highlights
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features:
- **Uniquely support of seamless switching between thinking mode** (for complex logical reasoning, math, and coding) and **non-thinking mode** (for efficient, general-purpose dialogue) **within single model**, ensuring optimal performance across various scenarios.
- **Significantly enhancement in its reasoning capabilities**, surpassing previous QwQ (in thinking mode) and Qwen2.5 instruct models (in non-thinking mode) on mathematics, code generation, and commonsense logical reasoning.
- **Superior human preference alignment**, excelling in creative writing, role-playing, multi-turn dialogues, and instruction following, to deliver a more natural, engaging, and immersive conversational experience.
- **Expertise in agent capabilities**, enabling precise integration with external tools in both thinking and unthinking modes and achieving leading performance among open-source models in complex agent-based tasks.
- **Support of 100+ languages and dialects** with strong capabilities for **multilingual instruction following** and **translation**.
## Model Overview
**Qwen3-32B** has the following features:
- Type: Causal Language Models
- Training Stage: Pretraining & Post-training
- Number of Parameters: 32.8B
- Number of Paramaters (Non-Embedding): 31.2B
- Number of Layers: 64
- Number of Attention Heads (GQA): 64 for Q and 8 for KV
- Context Length: 32,768 natively and [131,072 tokens with YaRN](#processing-long-texts).
For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our [blog](https://qwenlm.github.io/blog/qwen3/), [GitHub](https://github.com/QwenLM/Qwen3), and [Documentation](https://qwen.readthedocs.io/en/latest/).
## Quickstart
The code of Qwen3 has been in the latest Hugging Face `transformers` and we advise you to use the latest version of `transformers`.
With `transformers<4.51.0`, you will encounter the following error:
```
KeyError: 'qwen3'
```
The following contains a code snippet illustrating how to use the model generate content based on given inputs.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Qwen/Qwen3-32B"
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
# conduct text completion
generated_ids = model.generate(
**model_inputs,
max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
# parsing thinking content
try:
# rindex finding 151668 (</think>)
index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
index = 0
thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
print("thinking content:", thinking_content)
print("content:", content)
```
For deployment, you can use `sglang>=0.4.6.post1` or `vllm>=0.8.5` or to create an OpenAI-compatible API endpoint:
- SGLang:
```shell
python -m sglang.launch_server --model-path Qwen/Qwen3-32B --reasoning-parser qwen3
```
- vLLM:
```shell
vllm serve Qwen/Qwen3-32B --enable-reasoning --reasoning-parser deepseek_r1
```
For local use, applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers have also supported Qwen3.
## Switching Between Thinking and Non-Thinking Mode
> [!TIP]
> The `enable_thinking` switch is also available in APIs created by SGLang and vLLM.
> Please refer to our documentation for [SGLang](https://qwen.readthedocs.io/en/latest/deployment/sglang.html#thinking-non-thinking-modes) and [vLLM](https://qwen.readthedocs.io/en/latest/deployment/vllm.html#thinking-non-thinking-modes) users.
### `enable_thinking=True`
By default, Qwen3 has thinking capabilities enabled, similar to QwQ-32B. This means the model will use its reasoning abilities to enhance the quality of generated responses. For example, when explicitly setting `enable_thinking=True` or leaving it as the default value in `tokenizer.apply_chat_template`, the model will engage its thinking mode.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # True is the default value for enable_thinking
)
```
In this mode, the model will generate think content wrapped in a `<think>...</think>` block, followed by the final response.
> [!NOTE]
> For thinking mode, use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0` (the default setting in `generation_config.json`). **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### `enable_thinking=False`
We provide a hard switch to strictly disable the model's thinking behavior, aligning its functionality with the previous Qwen2.5-Instruct models. This mode is particularly useful in scenarios where disabling thinking is essential for enhancing efficiency.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False # Setting enable_thinking=False disables thinking mode
)
```
In this mode, the model will not generate any think content and will not include a `<think>...</think>` block.
> [!NOTE]
> For non-thinking mode, we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### Advanced Usage: Switching Between Thinking and Non-Thinking Modes via User Input
We provide a soft switch mechanism that allows users to dynamically control the model's behavior when `enable_thinking=True`. Specifically, you can add `/think` and `/no_think` to user prompts or system messages to switch the model's thinking mode from turn to turn. The model will follow the most recent instruction in multi-turn conversations.
Here is an example of a multi-turn conversation:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
class QwenChatbot:
def __init__(self, model_name="Qwen/Qwen3-32B"):
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
self.model = AutoModelForCausalLM.from_pretrained(model_name)
self.history = []
def generate_response(self, user_input):
messages = self.history + [{"role": "user", "content": user_input}]
text = self.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
inputs = self.tokenizer(text, return_tensors="pt")
response_ids = self.model.generate(**inputs, max_new_tokens=32768)[0][len(inputs.input_ids[0]):].tolist()
response = self.tokenizer.decode(response_ids, skip_special_tokens=True)
# Update history
self.history.append({"role": "user", "content": user_input})
self.history.append({"role": "assistant", "content": response})
return response
# Example Usage
if __name__ == "__main__":
chatbot = QwenChatbot()
# First input (without /think or /no_think tags, thinking mode is enabled by default)
user_input_1 = "How many r's in strawberries?"
print(f"User: {user_input_1}")
response_1 = chatbot.generate_response(user_input_1)
print(f"Bot: {response_1}")
print("----------------------")
# Second input with /no_think
user_input_2 = "Then, how many r's in blueberries? /no_think"
print(f"User: {user_input_2}")
response_2 = chatbot.generate_response(user_input_2)
print(f"Bot: {response_2}")
print("----------------------")
# Third input with /think
user_input_3 = "Really? /think"
print(f"User: {user_input_3}")
response_3 = chatbot.generate_response(user_input_3)
print(f"Bot: {response_3}")
```
> [!NOTE]
> For API compatibility, when `enable_thinking=True`, regardless of whether the user uses `/think` or `/no_think`, the model will always output a block wrapped in `<think>...</think>`. However, the content inside this block may be empty if thinking is disabled.
> When `enable_thinking=False`, the soft switches are not valid. Regardless of any `/think` or `/no_think` tags input by the user, the model will not generate think content and will not include a `<think>...</think>` block.
## Agentic Use
Qwen3 excels in tool calling capabilities. We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to make the best use of agentic ability of Qwen3. Qwen-Agent encapsulates tool-calling templates and tool-calling parsers internally, greatly reducing coding complexity.
To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
```python
from qwen_agent.agents import Assistant
# Define LLM
llm_cfg = {
'model': 'Qwen3-32B',
# Use the endpoint provided by Alibaba Model Studio:
# 'model_type': 'qwen_dashscope',
# 'api_key': os.getenv('DASHSCOPE_API_KEY'),
# Use a custom endpoint compatible with OpenAI API:
'model_server': 'http://localhost:8000/v1', # api_base
'api_key': 'EMPTY',
# Other parameters:
# 'generate_cfg': {
# # Add: When the response content is `<think>this is the thought</think>this is the answer;
# # Do not add: When the response has been separated by reasoning_content and content.
# 'thought_in_content': True,
# },
}
# Define Tools
tools = [
{'mcpServers': { # You can specify the MCP configuration file
'time': {
'command': 'uvx',
'args': ['mcp-server-time', '--local-timezone=Asia/Shanghai']
},
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
},
'code_interpreter', # Built-in tools
]
# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)
# Streaming generation
messages = [{'role': 'user', 'content': 'https://qwenlm.github.io/blog/ Introduce the latest developments of Qwen'}]
for responses in bot.run(messages=messages):
pass
print(responses)
```
## Processing Long Texts
Qwen3 natively supports context lengths of up to 32,768 tokens. For conversations where the total length (including both input and output) significantly exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively. We have validated the model's performance on context lengths of up to 131,072 tokens using the [YaRN](https://arxiv.org/abs/2309.00071) method.
YaRN is currently supported by several inference frameworks, e.g., `transformers` and `llama.cpp` for local use, `vllm` and `sglang` for deployment. In general, there are two approaches to enabling YaRN for supported frameworks:
- Modifying the model files:
In the `config.json` file, add the `rope_scaling` fields:
```json
{
...,
"rope_scaling": {
"rope_type": "yarn",
"factor": 4.0,
"original_max_position_embeddings": 32768
}
}
```
For `llama.cpp`, you need to regenerate the GGUF file after the modification.
- Passing command line arguments:
For `vllm`, you can use
```shell
vllm serve ... --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}' --max-model-len 131072
```
For `sglang`, you can use
```shell
python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling":{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}}'
```
For `llama-server` from `llama.cpp`, you can use
```shell
llama-server ... --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768
```
> [!IMPORTANT]
> If you encounter the following warning
> ```
> Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'original_max_position_embeddings'}
> ```
> please upgrade `transformers>=4.51.0`.
> [!NOTE]
> All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts.**
> We advise adding the `rope_scaling` configuration only when processing long contexts is required.
> It is also recommended to modify the `factor` as needed. For example, if the typical context length for your application is 65,536 tokens, it would be better to set `factor` as 2.0.
> [!NOTE]
> The default `max_position_embeddings` in `config.json` is set to 40,960. This allocation includes reserving 32,768 tokens for outputs and 8,192 tokens for typical prompts, which is sufficient for most scenarios involving short text processing. If the average context length does not exceed 32,768 tokens, we do not recommend enabling YaRN in this scenario, as it may potentially degrade model performance.
> [!TIP]
> The endpoint provided by Alibaba Model Studio supports dynamic YaRN by default and no extra configuration is needed.
## Best Practices
To achieve optimal performance, we recommend the following settings:
1. **Sampling Parameters**:
- For thinking mode (`enable_thinking=True`), use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0`. **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions.
- For non-thinking mode (`enable_thinking=False`), we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`.
- For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 38,912 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
- **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
- **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
4. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
### Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{qwen3,
title = {Qwen3},
url = {https://qwenlm.github.io/blog/qwen3/},
author = {Qwen Team},
month = {April},
year = {2025}
}
``` |
Mungert/openhands-lm-7b-v0.1-GGUF | Mungert | "2025-05-06T22:57:48Z" | 1,443 | 1 | null | [
"gguf",
"agent",
"coding",
"text-generation",
"en",
"dataset:SWE-Gym/SWE-Gym",
"arxiv:2412.21139",
"base_model:Qwen/Qwen2.5-Coder-7B-Instruct",
"base_model:quantized:Qwen/Qwen2.5-Coder-7B-Instruct",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-26T08:57:49Z" | ---
license: mit
datasets:
- SWE-Gym/SWE-Gym
language:
- en
base_model:
- Qwen/Qwen2.5-Coder-7B-Instruct
pipeline_tag: text-generation
tags:
- agent
- coding
---
# <span style="color: #7FFF7F;">openhands-lm-7b-v0.1 GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `openhands-lm-7b-v0.1-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `openhands-lm-7b-v0.1-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `openhands-lm-7b-v0.1-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `openhands-lm-7b-v0.1-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `openhands-lm-7b-v0.1-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `openhands-lm-7b-v0.1-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `openhands-lm-7b-v0.1-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `openhands-lm-7b-v0.1-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `openhands-lm-7b-v0.1-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `openhands-lm-7b-v0.1-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `openhands-lm-7b-v0.1-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
<div align="center">
<img src="https://github.com/All-Hands-AI/OpenHands/blob/main/docs/static/img/logo.png?raw=true" alt="Logo" width="200">
<h1 align="center">OpenHands LM v0.1</h1>
</div>
<p align="center">
<a href="https://www.all-hands.dev/blog/introducing-openhands-lm-32b----a-strong-open-coding-agent-model">Blog</a>
•
<a href="https://docs.all-hands.dev/modules/usage/llms/local-llms" >Use it in OpenHands</a>
</p>
**This is a smaller 7B model trained following the recipe of [all-hands/openhands-lm-32b-v0.1](https://huggingface.co/all-hands/openhands-lm-32b-v0.1).**
---
Autonomous agents for software development are already contributing to a [wide range of software development tasks](/blog/8-use-cases-for-generalist-software-development-agents).
But up to this point, strong coding agents have relied on proprietary models, which means that even if you use an open-source agent like [OpenHands](https://github.com/All-Hands-AI/OpenHands), you are still reliant on API calls to an external service.
Today, we are excited to introduce OpenHands LM, a new open coding model that:
- Is open and [available on Hugging Face](https://huggingface.co/all-hands/openhands-lm-32b-v0.1), so you can download it and run it locally
- Is a reasonable size, 32B, so it can be run locally on hardware such as a single 3090 GPU
- Achieves strong performance on software engineering tasks, including 37.2% resolve rate on SWE-Bench Verified
Read below for more details and our future plans!
## What is OpenHands LM?
OpenHands LM is built on the foundation of [Qwen Coder 2.5 Instruct 32B](https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct), leveraging its powerful base capabilities for coding tasks. What sets OpenHands LM apart is our specialized fine-tuning process:
- We used training data generated by OpenHands itself on a diverse set of open-source repositories
- Specifically, we use an RL-based framework outlined in [SWE-Gym](https://arxiv.org/abs/2412.21139), where we set up a training environment, generate training data using an existing agent, and then fine-tune the model on examples that were resolved successfully
- It features a 128K token context window, ideal for handling large codebases and long-horizon software engineering tasks
## Performance: Punching Above Its Weight
We evaluated OpenHands LM using our latest [iterative evaluation protocol](https://github.com/All-Hands-AI/OpenHands/tree/main/evaluation/benchmarks/swe_bench#run-inference-rollout-on-swe-bench-instances-generate-patch-from-problem-statement) on the [SWE-Bench Verified benchmark](https://www.swebench.com/#verified).
The results are impressive:
- **37.2% verified resolve rate** on SWE-Bench Verified
- Performance comparable to models with **20x more parameters**, including Deepseek V3 0324 (38.8%) with 671B parameters
Here's how OpenHands LM compares to other leading open-source models:
As the plot demonstrates, our 32B parameter model achieves efficiency that approaches much larger models. While the largest models (671B parameters) achieve slightly higher scores, our 32B parameter model performs remarkably well, opening up possibilities for local deployment that are not possible with larger models.
## Getting Started: How to Use OpenHands LM Today
You can start using OpenHands LM immediately through these channels:
1. **Download the model from Hugging Face**
The model is available on [Hugging Face](https://huggingface.co/all-hands/openhands-lm-32b-v0.1) and can be downloaded directly from there.
2. **Create an OpenAI-compatible endpoint with a model serving framework**
For optimal performance, it is recommended to serve this model with a GPU using [SGLang](https://github.com/sgl-project/sglang) or [vLLM](https://github.com/vllm-project/vllm).
3. **Point your OpenHands agent to the new model**
Download [OpenHands](https://github.com/All-Hands-AI/OpenHands) and follow the instructions for [using an OpenAI-compatible endpoint](https://docs.all-hands.dev/modules/usage/llms/openai-llms#using-openai-compatible-endpoints).
## The Road Ahead: Our Development Plans
This initial release marks just the beginning of our journey. We will continue enhancing OpenHands LM based on community feedback and ongoing research initiatives.
In particular, it should be noted that the model is still a research preview, and (1) may be best suited for tasks regarding solving github issues and perform less well on more varied software engineering tasks, (2) may sometimes generate repetitive steps, and (3) is somewhat sensitive to quantization, and may not function at full performance at lower quantization levels.
Our next releases will focus on addressing these limitations.
We're also developing more compact versions of the model (including a 7B parameter variant) to support users with limited computational resources. These smaller models will preserve OpenHands LM's core strengths while dramatically reducing hardware requirements.
We encourage you to experiment with OpenHands LM, share your experiences, and participate in its evolution. Together, we can create better tools for tomorrow's software development landscape.
## Try OpenHands Cloud
While OpenHands LM is a powerful model you can run locally, we also offer a fully managed cloud solution that makes it even easier to leverage AI for your software development needs.
[OpenHands Cloud](https://www.all-hands.dev/blog/introducing-the-openhands-cloud) provides:
- Seamless GitHub integration with issue and PR support
- Multiple interaction methods including text, voice, and mobile
- Parallel agent capabilities for working on multiple tasks simultaneously
- All the power of OpenHands without managing infrastructure
OpenHands Cloud is built on the same technology as our open-source solution but adds convenient features for teams and individuals who want a ready-to-use platform. [Visit app.all-hands.dev](https://app.all-hands.dev) to get started today!
## Join Our Community
We invite you to be part of the OpenHands LM journey:
- Explore our [GitHub repository](https://github.com/All-Hands-AI/OpenHands)
- Connect with us on [Slack](https://join.slack.com/t/openhands-ai/shared_invite/zt-2tom0er4l-JeNUGHt_AxpEfIBstbLPiw)
- Follow our [documentation](https://docs.all-hands.dev) to get started
By contributing your experiences and feedback, you'll help shape the future of this open-source initiative. Together, we can create better tools for tomorrow's software development landscape.
We can't wait to see what you'll create with OpenHands LM! |
Mungert/Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-GGUF | Mungert | "2025-05-06T22:57:13Z" | 1,525 | 3 | transformers | [
"transformers",
"gguf",
"en",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-22T19:51:00Z" | ---
library_name: transformers
language:
- en
license: cc-by-nc-4.0
---
# <span style="color: #7FFF7F;">Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct GGUF Models</span>
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Model Information
We introduce **Nemotron-UltraLong-8B**, a series of ultra-long context language models designed to process extensive sequences of text (up to 1M, 2M, and 4M tokens) while maintaining competitive performance on standard benchmarks. Built on the Llama-3.1, UltraLong-8B leverages a systematic training recipe that combines efficient continued pretraining with instruction tuning to enhance long-context understanding and instruction-following capabilities. This approach enables our models to efficiently scale their context windows without sacrificing general performance.
## The UltraLong Models
- [nvidia/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct](https://huggingface.co/nvidia/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct)
- [nvidia/Llama-3.1-Nemotron-8B-UltraLong-2M-Instruct](https://huggingface.co/nvidia/Llama-3.1-Nemotron-8B-UltraLong-2M-Instruct)
- [nvidia/Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct](https://huggingface.co/nvidia/Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct)
## Uses
Starting with `transformers >= 4.43.0` onward, you can run conversational inference using the Transformers `pipeline` abstraction or by leveraging the Auto classes with the `generate()` function.
Make sure to update your transformers installation via `pip install --upgrade transformers`.
```python
import transformers
import torch
model_id = "nvidia/Llama-3.1-Nemotron-8B-UltraLong-4M-Instruct"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
{"role": "user", "content": "Who are you?"},
]
outputs = pipeline(
messages,
max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])
```
## Model Card
* Base model: [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct)
* Continued Pretraining: The training data consists of 1B tokens sourced from a pretraining corpus using per-domain upsampling based on sample length. The model was trained for 150 iterations with a sequence length of 4M and a global batch size of 2.
* Supervised fine-tuning (SFT): 1B tokens on open-source instruction datasets across general, mathematics, and code domains. We subsample the data from the ‘general_sft_stage2’ from [AceMath-Instruct](https://huggingface.co/datasets/nvidia/AceMath-Instruct-Training-Data).
* Maximum context window: 4M tokens
## Evaluation Results
We evaluate Nemotron-UltraLong-8B on a diverse set of benchmarks, including long-context tasks (e.g., RULER, LV-Eval, and InfiniteBench) and standard tasks (e.g., MMLU, MATH, GSM-8K, and HumanEval). UltraLong-8B achieves superior performance on ultra-long context tasks while maintaining competitive results on standard benchmarks.
### Needle in a Haystack
<img width="80%" alt="image" src="Llama-3.1-8B-UltraLong-4M-Instruct.png">
### Long context evaluation
<img width="80%" alt="image" src="long_benchmark.png">
### Standard capability evaluation
<img width="80%" alt="image" src="standard_benchmark.png">
## Correspondence to
Chejian Xu ([email protected]), Wei Ping ([email protected])
## Citation
<pre>
@article{ulralong2025,
title={From 128K to 4M: Efficient Training of Ultra-Long Context Large Language Models},
author={Xu, Chejian and Ping, Wei and Xu, Peng and Liu, Zihan and Wang, Boxin and Shoeybi, Mohammad and Catanzaro, Bryan},
journal={arXiv preprint},
year={2025}
}
</pre> |
Mungert/granite-3.3-8b-instruct-GGUF | Mungert | "2025-05-06T22:57:09Z" | 906 | 3 | transformers | [
"transformers",
"gguf",
"language",
"granite-3.3",
"text-generation",
"arxiv:0000.00000",
"base_model:ibm-granite/granite-3.3-8b-base",
"base_model:quantized:ibm-granite/granite-3.3-8b-base",
"license:apache-2.0",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-17T17:50:12Z" | ---
pipeline_tag: text-generation
inference: false
license: apache-2.0
library_name: transformers
tags:
- language
- granite-3.3
base_model:
- ibm-granite/granite-3.3-8b-base
---
# <span style="color: #7FFF7F;">granite-3.3-8b-instruct GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `granite-3.3-8b-instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `granite-3.3-8b-instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `granite-3.3-8b-instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `granite-3.3-8b-instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `granite-3.3-8b-instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `granite-3.3-8b-instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `granite-3.3-8b-instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `granite-3.3-8b-instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `granite-3.3-8b-instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `granite-3.3-8b-instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `granite-3.3-8b-instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Granite-3.3-8B-Instruct
**Model Summary:**
Granite-3.3-8B-Instruct is a 8-billion parameter 128K context length language model fine-tuned for improved reasoning and instruction-following capabilities. Built on top of Granite-3.3-8B-Base, the model delivers significant gains on benchmarks for measuring generic performance including AlpacaEval-2.0 and Arena-Hard, and improvements in mathematics, coding, and instruction following. It supprts structured reasoning through \<think\>\<\/think\> and \<response\>\<\/response\> tags, providing clear separation between internal thoughts and final outputs. The model has been trained on a carefully balanced combination of permissively licensed data and curated synthetic tasks.
- **Developers:** Granite Team, IBM
- **Website**: [Granite Docs](https://www.ibm.com/granite/docs/)
- **Release Date**: April 16th, 2025
- **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
**Supported Languages:**
English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. However, users may finetune this Granite model for languages beyond these 12 languages.
**Intended Use:**
This model is designed to handle general instruction-following tasks and can be integrated into AI assistants across various domains, including business applications.
**Capabilities**
* Thinking
* Summarization
* Text classification
* Text extraction
* Question-answering
* Retrieval Augmented Generation (RAG)
* Code related tasks
* Function-calling tasks
* Multilingual dialog use cases
* Fill-in-the-middle
* Long-context tasks including long document/meeting summarization, long document QA, etc.
**Generation:**
This is a simple example of how to use Granite-3.3-8B-Instruct model.
Install the following libraries:
```shell
pip install torch torchvision torchaudio
pip install accelerate
pip install transformers
```
Then, copy the snippet from the section that is relevant for your use case.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
import torch
model_path="ibm-granite/granite-3.3-8b-instruct"
device="cuda"
model = AutoModelForCausalLM.from_pretrained(
model_path,
device_map=device,
torch_dtype=torch.bfloat16,
)
tokenizer = AutoTokenizer.from_pretrained(
model_path
)
conv = [{"role": "user", "content":"Redesign a common household item to make it more sustainable and user-friendly. Explain the changes and their benefits."}]
input_ids = tokenizer.apply_chat_template(conv, return_tensors="pt", thinking=True, return_dict=True, add_generation_prompt=True).to(device)
set_seed(42)
output = model.generate(
**input_ids,
max_new_tokens=8192,
)
prediction = tokenizer.decode(output[0, input_ids["input_ids"].shape[1]:], skip_special_tokens=True)
print(prediction)
```
**Example Outputs**
- thinking=True
```md
<think>To address this prompt, I'll focus on redesigning a common household item: the plastic water bottle. The goal is to make it more sustainable and user-friendly. Here are the proposed changes and their benefits:
1. **Material**: Replace plastic with biodegradable or recyclable materials like glass or plant-based plastics (PLA).
2. **Design**: Introduce a modular design for easy refilling and cleaning.
3. **Functionality**: Incorporate a built-in water filter for purification.
4. **Durability**: Enhance the bottle's lifespan through robust construction.
5. **User-friendliness**: Simplify the cap mechanism for easy opening and closing.
These changes would significantly reduce plastic waste, promote recycling, ensure access to clean drinking water, and enhance the overall user experience.
Now, I will detail each change and its benefits in a structured format.</think>
<response>### Redesign of the Plastic Water Bottle for Sustainability and User-Friendliness
#### 1. Material: Biodegradable or Recyclable Alternatives
**Change**: Transition from traditional PET plastic to glass or plant-based plastics (Polylactic Acid - PLA).
**Benefits**:
- **Sustainability**: Glass and PLA are biodegradable or highly recyclable, drastically reducing plastic pollution.
- **Health**: Unlike PET, these materials do not leach chemicals into the water, ensuring safer consumption.
#### 2. Modular Design for Easy Refilling and Cleaning
**Change**: Implement a detachable, interchangeable inner liner within a reusable outer casing.
**Benefits**:
- **User-Friendliness**: Users can easily remove the inner liner for thorough cleaning, reducing the risk of bacterial growth.
- **Sustainability**: The outer casing remains intact, encouraging reuse, while the inner liner can be replaced or recycled when worn out.
#### 3. Built-in Water Filter
**Change**: Integrate a removable, replaceable water filter into the bottle cap.
**Benefits**:
- **Accessibility**: Provides clean drinking water directly from any source (tap, river, etc.), eliminating the need for additional filtration devices.
- **Health**: Removes contaminants, bacteria, and impurities, ensuring safe hydration.
- **Sustainability**: Reduces the use of single-use plastic water bottles by offering a purification solution within a reusable container.
#### 4. Enhanced Durability
**Change**: Use high-quality, durable materials and robust construction techniques for both the outer casing and inner liner.
**Benefits**:
- **Longevity**: A more robust design increases the bottle's lifespan, reducing the frequency of replacements.
- **Cost-effectiveness**: Although the initial cost might be higher, the long-term savings from reduced replacements make it economical.
- **Sustainability**: Less frequent disposal means lower environmental impact.
#### 5. Simplified Cap Mechanism
**Change**: Adopt a one-touch, easy-to-open and close cap design, possibly inspired by sports bottle caps.
**Benefits**:
- **User-Friendliness**: Especially beneficial for children, elderly, or those with limited hand dexterity.
- **Convenience**: Quick access to water encourages more frequent hydration, promoting healthier habits.
### Conclusion
This redesigned water bottle addresses key sustainability concerns by minimizing plastic use and waste, while also enhancing user-friendliness through improved functionality, durability, and accessibility. By integrating biodegradable materials, a modular design, a built-in filter, robust construction, and a simple cap mechanism, the bottle not only reduces environmental footprint but also offers a superior user experience.</response>
```
- thinking=False
```md
Item for Redesign: Plastic Bottle (e.g., water or beverage bottle)
Redesign Concept: Eco-Friendly, Modular, and Multi-Functional Reusable Bottle
Changes:
1. Material: Replace the conventional plastic with food-grade, BPA-free, and recyclable materials such as stainless steel or glass. These materials are durable, non-toxic, and have a longer lifespan, reducing waste.
2. Modular Design: Introduce a modular design that allows users to customize their bottle according to their needs. This could include interchangeable caps, mouthpiece sizes, and carrying loops. Users can purchase additional modules as needed, promoting a more sustainable approach by reducing the need to buy an entirely new bottle for different purposes.
3. Integrated Filter: Incorporate a built-in, washable, and reusable filter that can remove impurities and improve the taste of water. This eliminates the need for single-use disposable filters or bottled water, further reducing plastic waste.
4. Smart Cap: Develop a smart cap with a built-in digital display and temperature sensor. This feature allows users to track their daily water intake, set hydration goals, and monitor the temperature of their beverage. The smart cap can be synced with a mobile app for additional functionality, such as reminders and progress tracking.
5. Easy-to-Clean Design: Ensure the bottle has a wide mouth and smooth interior surfaces for easy cleaning. Include a brush for hard-to-reach areas, making maintenance simple and encouraging regular use.
6. Collapsible Structure: Implement a collapsible design that reduces the bottle's volume when not in use, making it more portable and convenient for storage.
Benefits:
1. Sustainability: By using recyclable materials and reducing plastic waste, this redesigned bottle significantly contributes to a more sustainable lifestyle. The modular design and reusable filter also minimize single-use plastic consumption.
2. User-Friendly: The smart cap, easy-to-clean design, and collapsible structure make the bottle convenient and user-friendly. Users can customize their bottle to suit their needs, ensuring a better overall experience.
3. Healthier Option: Using food-grade, BPA-free materials and an integrated filter ensures that the beverages consumed are free from harmful chemicals and impurities, promoting a healthier lifestyle.
4. Cost-Effective: Although the initial investment might be higher, the long-term savings from reduced purchases of single-use plastic bottles and disposable filters make this reusable bottle a cost-effective choice.
5. Encourages Hydration: The smart cap's features, such as hydration tracking and temperature monitoring, can motivate users to stay hydrated and develop healthier habits.
By redesigning a common household item like the plastic bottle, we can create a more sustainable, user-friendly, and health-conscious alternative that benefits both individuals and the environment.
```
**Evaluation Results:**
<table>
<thead>
<caption style="text-align:center"><b>Comparison with different models over various benchmarks<sup id="fnref1"><a href="#fn1">1</a></sup>. Scores of AlpacaEval-2.0 and Arena-Hard are calculated with thinking=True</b></caption>
<tr>
<th style="text-align:left; background-color: #001d6c; color: white;">Models</th>
<th style="text-align:center; background-color: #001d6c; color: white;">Arena-Hard</th>
<th style="text-align:center; background-color: #001d6c; color: white;">AlpacaEval-2.0</th>
<th style="text-align:center; background-color: #001d6c; color: white;">MMLU</th>
<th style="text-align:center; background-color: #001d6c; color: white;">PopQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">TruthfulQA</th>
<th style="text-align:center; background-color: #001d6c; color: white;">BigBenchHard<sup id="fnref2"><a href="#fn2">2</a></sup></th>
<th style="text-align:center; background-color: #001d6c; color: white;">DROP<sup id="fnref3"><a href="#fn3">3</a></sup></th>
<th style="text-align:center; background-color: #001d6c; color: white;">GSM8K</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">HumanEval+</th>
<th style="text-align:center; background-color: #001d6c; color: white;">IFEval</th>
<th style="text-align:center; background-color: #001d6c; color: white;">AttaQ</th>
</tr></thead>
<tbody>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.1-2B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">23.3</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">27.17</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">57.11</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">20.55</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">59.79</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">61.82</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">20.99</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">67.55</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">79.45</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">75.26</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">63.59</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">84.7</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.2-2B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">24.86</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">34.51</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">57.18</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">20.56</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">59.8</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">61.39</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">23.84</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">67.02</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">80.13</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">73.39</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">61.55</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">83.23</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.3-2B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 28.86 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 43.45 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 55.88 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 18.4 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 58.97 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 63.91 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 44.33 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 72.48 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 80.51 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 75.68 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 65.8 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">87.47</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Llama-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">36.43</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">27.22</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">69.15</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">28.79</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">52.79</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">73.43</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">71.23</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">83.24</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">85.32</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">80.15</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">79.10</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">83.43</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">DeepSeek-R1-Distill-Llama-8B</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">17.17</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">21.85</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">45.80</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">13.25</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">47.43</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">67.39</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">49.73</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">72.18</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">67.54</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">62.91</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">66.50</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">42.87</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Qwen-2.5-7B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">25.44</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">30.34</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">74.30</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">18.12</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">63.06</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">69.19</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">64.06</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">84.46</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">93.35</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">89.91</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">74.90</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">81.90</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">DeepSeek-R1-Distill-Qwen-7B</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">10.36</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">15.35</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">50.72</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">9.94</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">47.14</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">67.38</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">51.78</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">78.47</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">79.89</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">78.43</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">59.10</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">42.45</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">37.58</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">30.34</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">66.77</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">28.7</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">65.84</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">69.87</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">58.57</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">79.15</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">89.63</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">85.79</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">73.20</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">85.73</td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.2-8B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">55.25</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">61.19</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">66.79</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">28.04</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">66.92</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">71.86</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">58.29</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">81.65</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">89.35</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">85.72</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">74.31</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">84.7</td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.3-8B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 57.56 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 62.68 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 65.54 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 26.17 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 66.86 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 69.13 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 59.36 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 80.89 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 89.73 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 86.09 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 74.82 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;">88.5</td>
</tr>
</tbody></table>
<table>
<caption style="text-align:center"><b>Math Benchmarks</b></caption>
<thead>
<tr>
<th style="text-align:left; background-color: #001d6c; color: white;">Models</th>
<th style="text-align:center; background-color: #001d6c; color: white;">AIME24</th>
<th style="text-align:center; background-color: #001d6c; color: white;">MATH-500</th>
</tr></thead>
<tbody>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.1-2B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 0.89 </td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 35.07 </td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.2-2B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 0.89 </td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 35.54 </td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.3-2B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 3.28 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 58.09 </td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.1-8B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 1.97 </td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 48.73 </td>
</tr>
<tr>
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.2-8B-Instruct</td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 2.43 </td>
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;"> 52.8 </td>
</tr>
<tr>
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.3-8B-Instruct</b></td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 8.12 </td>
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 69.02 </td>
</tr>
</tbody></table>
**Training Data:**
Overall, our training data is largely comprised of two key sources: (1) publicly available datasets with permissive license, (2) internal synthetically generated data targeted to enhance reasoning capabilites.
<!-- A detailed attribution of datasets can be found in [Granite 3.2 Technical Report (coming soon)](#), and [Accompanying Author List](https://github.com/ibm-granite/granite-3.0-language-models/blob/main/author-ack.pdf). -->
**Infrastructure:**
We train Granite-3.3-8B-Instruct using IBM's super computing cluster, Blue Vela, which is outfitted with NVIDIA H100 GPUs. This cluster provides a scalable and efficient infrastructure for training our models over thousands of GPUs.
**Ethical Considerations and Limitations:**
Granite-3.3-8B-Instruct builds upon Granite-3.3-8B-Base, leveraging both permissively licensed open-source and select proprietary data for enhanced performance. Since it inherits its foundation from the previous model, all ethical considerations and limitations applicable to [Granite-3.3-8B-Base](https://huggingface.co/ibm-granite/granite-3.3-8b-base) remain relevant.
**Resources**
- ⭐️ Learn about the latest updates with Granite: https://www.ibm.com/granite
- 📄 Get started with tutorials, best practices, and prompt engineering advice: https://www.ibm.com/granite/docs/
- 💡 Learn about the latest Granite learning resources: https://ibm.biz/granite-learning-resources
<p><a href="#fnref1" title="Jump back to reference">[1]</a> Evaluated using <a href="https://github.com/allenai/olmes">OLMES</a> (except AttaQ and Arena-Hard scores)</p>
<p><a href="#fnref2" title="Jump back to reference">[2]</a> Added regex for more efficient asnwer extraction.</a></p>
<p><a href="#fnref3" title="Jump back to reference">[3]</a> Modified the implementation to handle some of the issues mentioned <a href="https://huggingface.co/blog/open-llm-leaderboard-drop">here</a></p>
<!-- ## Citation
<!-- ## Citation
```
@misc{granite-models,
author = {author 1, author2, ...},
title = {},
journal = {},
volume = {},
year = {2024},
url = {https://arxiv.org/abs/0000.00000},
}
``` --> |
Mungert/watt-tool-8B-GGUF | Mungert | "2025-05-06T22:57:00Z" | 1,527 | 3 | null | [
"gguf",
"function-calling",
"tool-use",
"llama",
"bfcl",
"en",
"arxiv:2406.14868",
"base_model:meta-llama/Llama-3.1-8B-Instruct",
"base_model:quantized:meta-llama/Llama-3.1-8B-Instruct",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-15T01:43:23Z" | ---
license: apache-2.0
language:
- en
base_model:
- meta-llama/Llama-3.1-8B-Instruct
tags:
- function-calling
- tool-use
- llama
- bfcl
---
# <span style="color: #7FFF7F;">watt-tool-8B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `watt-tool-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `watt-tool-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `watt-tool-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `watt-tool-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `watt-tool-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `watt-tool-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `watt-tool-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `watt-tool-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `watt-tool-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `watt-tool-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `watt-tool-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# watt-tool-8B
watt-tool-8B is a fine-tuned language model based on LLaMa-3.1-8B-Instruct, optimized for tool usage and multi-turn dialogue. It achieves state-of-the-art performance on the Berkeley Function-Calling Leaderboard (BFCL).
## Model Description
This model is specifically designed to excel at complex tool usage scenarios that require multi-turn interactions, making it ideal for empowering platforms like [Lupan](https://lupan.watt.chat), an AI-powered workflow building tool. By leveraging a carefully curated and optimized dataset, watt-tool-8B demonstrates superior capabilities in understanding user requests, selecting appropriate tools, and effectively utilizing them across multiple turns of conversation.
Target Application: AI Workflow Building as in [https://lupan.watt.chat/](https://lupan.watt.chat/) and [Coze](https://www.coze.com/).
## Key Features
* **Enhanced Tool Usage:** Fine-tuned for precise and efficient tool selection and execution.
* **Multi-Turn Dialogue:** Optimized for maintaining context and effectively utilizing tools across multiple turns of conversation, enabling more complex task completion.
* **State-of-the-Art Performance:** Achieves top performance on the BFCL, demonstrating its capabilities in function calling and tool usage.
## Training Methodology
watt-tool-8B is trained using supervised fine-tuning on a specialized dataset designed for tool usage and multi-turn dialogue. We use CoT techniques to synthesize high-quality multi-turn dialogue data.
The training process is inspired by the principles outlined in the paper: ["Direct Multi-Turn Preference Optimization for Language Agents"](https://arxiv.org/abs/2406.14868).
We use SFT and DMPO to further enhance the model's performance in multi-turn agent tasks.
## How to Use
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "watt-ai/watt-tool-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype='auto', device_map="auto")
# Example usage (adapt as needed for your specific tool usage scenario)
"""You are an expert in composing functions. You are given a question and a set of possible functions. Based on the question, you will need to make one or more function/tool calls to achieve the purpose.
If none of the function can be used, point it out. If the given question lacks the parameters required by the function, also point it out.
You should only return the function call in tools call sections.
If you decide to invoke any of the function(s), you MUST put it in the format of [func_name1(params_name1=params_value1, params_name2=params_value2...), func_name2(params)]
You SHOULD NOT include any other text in the response.
Here is a list of functions in JSON format that you can invoke.\n{functions}\n
"""
# User query
query = "Find me the sales growth rate for company XYZ for the last 3 years and also the interest coverage ratio for the same duration."
tools = [
{
"name": "financial_ratios.interest_coverage", "description": "Calculate a company's interest coverage ratio given the company name and duration",
"arguments": {
"type": "dict",
"properties": {
"company_name": {
"type": "string",
"description": "The name of the company."
},
"years": {
"type": "integer",
"description": "Number of past years to calculate the ratio."
}
},
"required": ["company_name", "years"]
}
},
{
"name": "sales_growth.calculate",
"description": "Calculate a company's sales growth rate given the company name and duration",
"arguments": {
"type": "dict",
"properties": {
"company": {
"type": "string",
"description": "The company that you want to get the sales growth rate for."
},
"years": {
"type": "integer",
"description": "Number of past years for which to calculate the sales growth rate."
}
},
"required": ["company", "years"]
}
},
{
"name": "weather_forecast",
"description": "Retrieve a weather forecast for a specific location and time frame.",
"arguments": {
"type": "dict",
"properties": {
"location": {
"type": "string",
"description": "The city that you want to get the weather for."
},
"days": {
"type": "integer",
"description": "Number of days for the forecast."
}
},
"required": ["location", "days"]
}
}
]
messages = [
{'role': 'system', 'content': system_prompt.format(functions=tools)},
{'role': 'user', 'content': query}
]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)) |
Mungert/Llama-3.1-Nemotron-70B-Instruct-HF-GGUF | Mungert | "2025-05-06T22:56:47Z" | 1,611 | 4 | transformers | [
"transformers",
"gguf",
"nvidia",
"llama3.1",
"text-generation",
"en",
"dataset:nvidia/HelpSteer2",
"arxiv:2410.01257",
"arxiv:2405.01481",
"arxiv:2406.08673",
"base_model:meta-llama/Llama-3.1-70B-Instruct",
"base_model:quantized:meta-llama/Llama-3.1-70B-Instruct",
"license:llama3.1",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-06T22:54:06Z" | ---
license: llama3.1
language:
- en
inference: false
fine-tuning: false
tags:
- nvidia
- llama3.1
datasets:
- nvidia/HelpSteer2
base_model: meta-llama/Llama-3.1-70B-Instruct
pipeline_tag: text-generation
library_name: transformers
---
# <span style="color: #7FFF7F;">Llama-3.1-Nemotron-70B-Instruct-HF GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.1-Nemotron-70B-Instruct-HF-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.1-Nemotron-70B-Instruct-HF-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.1-Nemotron-70B-Instruct-HF-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.1-Nemotron-70B-Instruct-HF-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.1-Nemotron-70B-Instruct-HF-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.1-Nemotron-70B-Instruct-HF-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.1-Nemotron-70B-Instruct-HF-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.1-Nemotron-70B-Instruct-HF-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.1-Nemotron-70B-Instruct-HF-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.1-Nemotron-70B-Instruct-HF-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.1-Nemotron-70B-Instruct-HF-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Model Overview
## Description:
Llama-3.1-Nemotron-70B-Instruct is a large language model customized by NVIDIA to improve the helpfulness of LLM generated responses to user queries.
This model reaches [Arena Hard](https://github.com/lmarena/arena-hard-auto) of 85.0, [AlpacaEval 2 LC](https://tatsu-lab.github.io/alpaca_eval/) of 57.6 and [GPT-4-Turbo MT-Bench](https://github.com/lm-sys/FastChat/pull/3158) of 8.98, which are known to be predictive of [LMSys Chatbot Arena Elo](https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard)
As of 1 Oct 2024, this model is #1 on all three automatic alignment benchmarks (verified tab for AlpacaEval 2 LC), edging out strong frontier models such as GPT-4o and Claude 3.5 Sonnet.
As of Oct 24th, 2024 the model has Elo Score of 1267(+-7), rank 9 and style controlled rank of 26 on [ChatBot Arena leaderboard](https://lmarena.ai/?leaderboard).
This model was trained using RLHF (specifically, REINFORCE), [Llama-3.1-Nemotron-70B-Reward](https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Reward) and [HelpSteer2-Preference prompts](https://huggingface.co/datasets/nvidia/HelpSteer2) on a Llama-3.1-70B-Instruct model as the initial policy.
Llama-3.1-Nemotron-70B-Instruct-HF has been converted from [Llama-3.1-Nemotron-70B-Instruct](https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct) to support it in the HuggingFace Transformers codebase. Please note that evaluation results might be slightly different from the [Llama-3.1-Nemotron-70B-Instruct](https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct) as evaluated in NeMo-Aligner, which the evaluation results below are based on.
Try hosted inference for free at [build.nvidia.com](https://build.nvidia.com/nvidia/llama-3_1-nemotron-70b-instruct) - it comes with an OpenAI-compatible API interface.
See details on our paper at [https://arxiv.org/abs/2410.01257](https://arxiv.org/abs/2410.01257) - as a preview, this model can correctly the question ```How many r in strawberry?``` without specialized prompting or additional reasoning tokens:
```
A sweet question!
Let’s count the “R”s in “strawberry”:
1. S
2. T
3. R
4. A
5. W
6. B
7. E
8. R
9. R
10. Y
There are **3 “R”s** in the word “strawberry”.
```
Note: This model is a demonstration of our techniques for improving helpfulness in general-domain instruction following. It has not been tuned for performance in specialized domains such as math.
## Terms of use
By accessing this model, you are agreeing to the LLama 3.1 terms and conditions of the [license](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE), [acceptable use policy](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/USE_POLICY.md) and [Meta’s privacy policy](https://www.facebook.com/privacy/policy/)
## Evaluation Metrics
As of 1 Oct 2024, Llama-3.1-Nemotron-70B-Instruct performs best on Arena Hard, AlpacaEval 2 LC (verified tab) and MT Bench (GPT-4-Turbo)
| Model | Arena Hard | AlpacaEval | MT-Bench | Mean Response Length |
|:-----------------------------|:----------------|:-----|:----------|:-------|
|Details | (95% CI) | 2 LC (SE) | (GPT-4-Turbo) | (# of Characters for MT-Bench)|
| _**Llama-3.1-Nemotron-70B-Instruct**_ | **85.0** (-1.5, 1.5) | **57.6** (1.65) | **8.98** | 2199.8 |
| Llama-3.1-70B-Instruct | 55.7 (-2.9, 2.7) | 38.1 (0.90) | 8.22 | 1728.6 |
| Llama-3.1-405B-Instruct | 69.3 (-2.4, 2.2) | 39.3 (1.43) | 8.49 | 1664.7 |
| Claude-3-5-Sonnet-20240620 | 79.2 (-1.9, 1.7) | 52.4 (1.47) | 8.81 | 1619.9 |
| GPT-4o-2024-05-13 | 79.3 (-2.1, 2.0) | 57.5 (1.47) | 8.74 | 1752.2 |
## Usage:
You can use the model using HuggingFace Transformers library with 2 or more 80GB GPUs (NVIDIA Ampere or newer) with at least 150GB of free disk space to accomodate the download.
This code has been tested on Transformers v4.44.0, torch v2.4.0 and 2 A100 80GB GPUs, but any setup that supports ```meta-llama/Llama-3.1-70B-Instruct``` should support this model as well. If you run into problems, you can consider doing ```pip install -U transformers```.
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "nvidia/Llama-3.1-Nemotron-70B-Instruct-HF"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "How many r in strawberry?"
messages = [{"role": "user", "content": prompt}]
tokenized_message = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt", return_dict=True)
response_token_ids = model.generate(tokenized_message['input_ids'].cuda(),attention_mask=tokenized_message['attention_mask'].cuda(), max_new_tokens=4096, pad_token_id = tokenizer.eos_token_id)
generated_tokens =response_token_ids[:, len(tokenized_message['input_ids'][0]):]
generated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
print(generated_text)
# See response at top of model card
```
## References(s):
* [NeMo Aligner](https://arxiv.org/abs/2405.01481)
* [HelpSteer2-Preference](https://arxiv.org/abs/2410.01257)
* [HelpSteer2](https://arxiv.org/abs/2406.08673)
* [Introducing Llama 3.1: Our most capable models to date](https://ai.meta.com/blog/meta-llama-3-1/)
* [Meta's Llama 3.1 Webpage](https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_1)
* [Meta's Llama 3.1 Model Card](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md)
## Model Architecture:
**Architecture Type:** Transformer <br>
**Network Architecture:** Llama 3.1 <br>
## Input:
**Input Type(s):** Text <br>
**Input Format:** String <br>
**Input Parameters:** One Dimensional (1D) <br>
**Other Properties Related to Input:** Max of 128k tokens<br>
## Output:
**Output Type(s):** Text <br>
**Output Format:** String <br>
**Output Parameters:** One Dimensional (1D) <br>
**Other Properties Related to Output:** Max of 4k tokens <br>
## Software Integration:
**Supported Hardware Microarchitecture Compatibility:** <br>
* NVIDIA Ampere <br>
* NVIDIA Hopper <br>
* NVIDIA Turing <br>
**Supported Operating System(s):** Linux <br>
## Model Version:
v1.0
# Training & Evaluation:
## Alignment methodology
* REINFORCE implemented in NeMo Aligner
## Datasets:
**Data Collection Method by dataset** <br>
* [Hybrid: Human, Synthetic] <br>
**Labeling Method by dataset** <br>
* [Human] <br>
**Link:**
* [HelpSteer2](https://huggingface.co/datasets/nvidia/HelpSteer2)
**Properties (Quantity, Dataset Descriptions, Sensor(s)):** <br>
* 21, 362 prompt-responses built to make more models more aligned with human preference - specifically more helpful, factually-correct, coherent, and customizable based on complexity and verbosity.
* 20, 324 prompt-responses used for training and 1, 038 used for validation.
# Inference:
**Engine:** [Triton](https://developer.nvidia.com/triton-inference-server) <br>
**Test Hardware:** H100, A100 80GB, A100 40GB <br>
## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
## Citation
If you find this model useful, please cite the following works
```bibtex
@misc{wang2024helpsteer2preferencecomplementingratingspreferences,
title={HelpSteer2-Preference: Complementing Ratings with Preferences},
author={Zhilin Wang and Alexander Bukharin and Olivier Delalleau and Daniel Egert and Gerald Shen and Jiaqi Zeng and Oleksii Kuchaiev and Yi Dong},
year={2024},
eprint={2410.01257},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2410.01257},
}
``` |
Mungert/QwQ-32B-GGUF | Mungert | "2025-05-06T22:56:43Z" | 6,686 | 18 | transformers | [
"transformers",
"gguf",
"chat",
"text-generation",
"zho",
"eng",
"fra",
"spa",
"por",
"deu",
"ita",
"rus",
"jpn",
"kor",
"vie",
"tha",
"ara",
"arxiv:2309.00071",
"arxiv:2412.15115",
"base_model:Qwen/Qwen2.5-32B",
"base_model:quantized:Qwen/Qwen2.5-32B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-04T22:21:28Z" | ---
license: apache-2.0
license_link: https://huggingface.co/Qwen/QWQ-32B/blob/main/LICENSE
language:
- zho
- eng
- fra
- spa
- por
- deu
- ita
- rus
- jpn
- kor
- vie
- tha
- ara
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-32B
tags:
- chat
library_name: transformers
---
# <span style="color: #7FFF7F;">QwQ-32B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `QwQ-32B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `QwQ-32B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `QwQ-32B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `QwQ-32B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `QwQ-32B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `QwQ-32B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `QwQ-32B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `QwQ-32B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `QwQ-32B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `QwQ-32B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `QwQ-32B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# QwQ-32B
<a href="https://chat.qwenlm.ai/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
</a>
## Introduction
QwQ is the reasoning model of the Qwen series. Compared with conventional instruction-tuned models, QwQ, which is capable of thinking and reasoning, can achieve significantly enhanced performance in downstream tasks, especially hard problems. QwQ-32B is the medium-sized reasoning model, which is capable of achieving competitive performance against state-of-the-art reasoning models, e.g., DeepSeek-R1, o1-mini.
<p align="center">
<img width="100%" src="figures/benchmark.jpg">
</p>
**This repo contains the QwQ 32B model**, which has the following features:
- Type: Causal Language Models
- Training Stage: Pretraining & Post-training (Supervised Finetuning and Reinforcement Learning)
- Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias
- Number of Parameters: 32.5B
- Number of Paramaters (Non-Embedding): 31.0B
- Number of Layers: 64
- Number of Attention Heads (GQA): 40 for Q and 8 for KV
- Context Length: Full 131,072 tokens
- For prompts exceeding 8,192 tokens in length, you must enable YaRN as outlined in [this section](#usage-guidelines).
**Note:** For the best experience, please review the [usage guidelines](#usage-guidelines) before deploying QwQ models.
You can try our [demo](https://huggingface.co/spaces/Qwen/QwQ-32B-Demo) or access QwQ models via [QwenChat](https://chat.qwen.ai).
For more details, please refer to our [blog](https://qwenlm.github.io/blog/qwq-32b/), [GitHub](https://github.com/QwenLM/Qwen2.5), and [Documentation](https://qwen.readthedocs.io/en/latest/).
## Requirements
QwQ is based on Qwen2.5, whose code has been in the latest Hugging face `transformers`. We advise you to use the latest version of `transformers`.
With `transformers<4.37.0`, you will encounter the following error:
```
KeyError: 'qwen2'
```
## Quickstart
Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Qwen/QwQ-32B"
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "How many r's are in the word \"strawberry\""
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=32768
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
```
### Usage Guidelines
To achieve optimal performance, we recommend the following settings:
1. **Enforce Thoughtful Output**: Ensure the model starts with "\<think\>\n" to prevent generating empty thinking content, which can degrade output quality. If you use `apply_chat_template` and set `add_generation_prompt=True`, this is already automatically implemented, but it may cause the response to lack the \<think\> tag at the beginning. This is normal behavior.
2. **Sampling Parameters**:
- Use Temperature=0.6, TopP=0.95, MinP=0 instead of Greedy decoding to avoid endless repetitions.
- Use TopK between 20 and 40 to filter out rare token occurrences while maintaining the diversity of the generated output.
- For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may result in occasional language mixing and a slight decrease in performance.
3. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. This feature is already implemented in `apply_chat_template`.
4. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
- **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
- **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g.,`\"answer\": \"C\"`." in the prompt.
5. **Handle Long Inputs**: For inputs exceeding 8,192 tokens, enable [YaRN](https://arxiv.org/abs/2309.00071) to improve the model's ability to capture long-sequence information effectively.
For supported frameworks, you could add the following to `config.json` to enable YaRN:
```json
{
...,
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
}
```
For deployment, we recommend using vLLM. Please refer to our [Documentation](https://qwen.readthedocs.io/en/latest/deployment/vllm.html) for usage if you are not familar with vLLM.
Presently, vLLM only supports static YARN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts**.
We advise adding the `rope_scaling` configuration only when processing long contexts is required.
## Evaluation & Performance
Detailed evaluation results are reported in this [📑 blog](https://qwenlm.github.io/blog/qwq-32b/).
For requirements on GPU memory and the respective throughput, see results [here](https://qwen.readthedocs.io/en/latest/benchmark/speed_benchmark.html).
## Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{qwq32b,
title = {QwQ-32B: Embracing the Power of Reinforcement Learning},
url = {https://qwenlm.github.io/blog/qwq-32b/},
author = {Qwen Team},
month = {March},
year = {2025}
}
@article{qwen2.5,
title={Qwen2.5 Technical Report},
author={An Yang and Baosong Yang and Beichen Zhang and Binyuan Hui and Bo Zheng and Bowen Yu and Chengyuan Li and Dayiheng Liu and Fei Huang and Haoran Wei and Huan Lin and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Yang and Jiaxi Yang and Jingren Zhou and Junyang Lin and Kai Dang and Keming Lu and Keqin Bao and Kexin Yang and Le Yu and Mei Li and Mingfeng Xue and Pei Zhang and Qin Zhu and Rui Men and Runji Lin and Tianhao Li and Tianyi Tang and Tingyu Xia and Xingzhang Ren and Xuancheng Ren and Yang Fan and Yang Su and Yichang Zhang and Yu Wan and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zihan Qiu},
journal={arXiv preprint arXiv:2412.15115},
year={2024}
}
``` |
Mungert/DeepSeek-R1-Distill-Qwen-32B-GGUF | Mungert | "2025-05-06T22:56:32Z" | 18,366 | 4 | transformers | [
"transformers",
"gguf",
"arxiv:2501.12948",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-03T12:11:49Z" | ---
license: mit
library_name: transformers
---
# <span style="color: #7FFF7F;">DeepSeek-R1-Distill-Qwen-32B GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `DeepSeek-R1-Distill-Qwen-32B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `DeepSeek-R1-Distill-Qwen-32B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `DeepSeek-R1-Distill-Qwen-32B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `DeepSeek-R1-Distill-Qwen-32B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `DeepSeek-R1-Distill-Qwen-32B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `DeepSeek-R1-Distill-Qwen-32B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `DeepSeek-R1-Distill-Qwen-32B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `DeepSeek-R1-Distill-Qwen-32B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `DeepSeek-R1-Distill-Qwen-32B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `DeepSeek-R1-Distill-Qwen-32B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `DeepSeek-R1-Distill-Qwen-32B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# DeepSeek-R1
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<!-- markdownlint-disable no-duplicate-header -->
<div align="center">
<img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V3" />
</div>
<hr>
<div align="center" style="line-height: 1;">
<a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20R1-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://discord.gg/Tc7c45Zzu5" target="_blank" style="margin: 2px;">
<img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/qr.jpeg?raw=true" target="_blank" style="margin: 2px;">
<img alt="Wechat" src="https://img.shields.io/badge/WeChat-DeepSeek%20AI-brightgreen?logo=wechat&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
<a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;">
<img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<div align="center" style="line-height: 1;">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE" style="margin: 2px;">
<img alt="License" src="https://img.shields.io/badge/License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>
</a>
</div>
<p align="center">
<a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>
</p>
## 1. Introduction
We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.
DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning.
With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.
However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,
we introduce DeepSeek-R1, which incorporates cold-start data before RL.
DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.
To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.
**NOTE: Before running DeepSeek-R1 series models locally, we kindly recommend reviewing the [Usage Recommendation](#usage-recommendations) section.**
<p align="center">
<img width="80%" src="figures/benchmark.jpg">
</p>
## 2. Model Summary
---
**Post-Training: Large-Scale Reinforcement Learning on the Base Model**
- We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.
- We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.
We believe the pipeline will benefit the industry by creating better models.
---
**Distillation: Smaller Models Can Be Powerful Too**
- We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future.
- Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.
## 3. Model Downloads
### DeepSeek-R1 Models
<div align="center">
| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
| :------------: | :------------: | :------------: | :------------: | :------------: |
| DeepSeek-R1-Zero | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero) |
| DeepSeek-R1 | 671B | 37B | 128K | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1) |
</div>
DeepSeek-R1-Zero & DeepSeek-R1 are trained based on DeepSeek-V3-Base.
For more details regarding the model architecture, please refer to [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repository.
### DeepSeek-R1-Distill Models
<div align="center">
| **Model** | **Base Model** | **Download** |
| :------------: | :------------: | :------------: |
| DeepSeek-R1-Distill-Qwen-1.5B | [Qwen2.5-Math-1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) |
| DeepSeek-R1-Distill-Qwen-7B | [Qwen2.5-Math-7B](https://huggingface.co/Qwen/Qwen2.5-Math-7B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B) |
| DeepSeek-R1-Distill-Llama-8B | [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B) |
| DeepSeek-R1-Distill-Qwen-14B | [Qwen2.5-14B](https://huggingface.co/Qwen/Qwen2.5-14B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B) |
|DeepSeek-R1-Distill-Qwen-32B | [Qwen2.5-32B](https://huggingface.co/Qwen/Qwen2.5-32B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) |
| DeepSeek-R1-Distill-Llama-70B | [Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B) |
</div>
DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1.
We slightly change their configs and tokenizers. Please use our setting to run these models.
## 4. Evaluation Results
### DeepSeek-R1-Evaluation
For all our models, the maximum generation length is set to 32,768 tokens. For benchmarks requiring sampling, we use a temperature of $0.6$, a top-p value of $0.95$, and generate 64 responses per query to estimate pass@1.
<div align="center">
| Category | Benchmark (Metric) | Claude-3.5-Sonnet-1022 | GPT-4o 0513 | DeepSeek V3 | OpenAI o1-mini | OpenAI o1-1217 | DeepSeek R1 |
|----------|-------------------|----------------------|------------|--------------|----------------|------------|--------------|
| | Architecture | - | - | MoE | - | - | MoE |
| | # Activated Params | - | - | 37B | - | - | 37B |
| | # Total Params | - | - | 671B | - | - | 671B |
| English | MMLU (Pass@1) | 88.3 | 87.2 | 88.5 | 85.2 | **91.8** | 90.8 |
| | MMLU-Redux (EM) | 88.9 | 88.0 | 89.1 | 86.7 | - | **92.9** |
| | MMLU-Pro (EM) | 78.0 | 72.6 | 75.9 | 80.3 | - | **84.0** |
| | DROP (3-shot F1) | 88.3 | 83.7 | 91.6 | 83.9 | 90.2 | **92.2** |
| | IF-Eval (Prompt Strict) | **86.5** | 84.3 | 86.1 | 84.8 | - | 83.3 |
| | GPQA-Diamond (Pass@1) | 65.0 | 49.9 | 59.1 | 60.0 | **75.7** | 71.5 |
| | SimpleQA (Correct) | 28.4 | 38.2 | 24.9 | 7.0 | **47.0** | 30.1 |
| | FRAMES (Acc.) | 72.5 | 80.5 | 73.3 | 76.9 | - | **82.5** |
| | AlpacaEval2.0 (LC-winrate) | 52.0 | 51.1 | 70.0 | 57.8 | - | **87.6** |
| | ArenaHard (GPT-4-1106) | 85.2 | 80.4 | 85.5 | 92.0 | - | **92.3** |
| Code | LiveCodeBench (Pass@1-COT) | 33.8 | 34.2 | - | 53.8 | 63.4 | **65.9** |
| | Codeforces (Percentile) | 20.3 | 23.6 | 58.7 | 93.4 | **96.6** | 96.3 |
| | Codeforces (Rating) | 717 | 759 | 1134 | 1820 | **2061** | 2029 |
| | SWE Verified (Resolved) | **50.8** | 38.8 | 42.0 | 41.6 | 48.9 | 49.2 |
| | Aider-Polyglot (Acc.) | 45.3 | 16.0 | 49.6 | 32.9 | **61.7** | 53.3 |
| Math | AIME 2024 (Pass@1) | 16.0 | 9.3 | 39.2 | 63.6 | 79.2 | **79.8** |
| | MATH-500 (Pass@1) | 78.3 | 74.6 | 90.2 | 90.0 | 96.4 | **97.3** |
| | CNMO 2024 (Pass@1) | 13.1 | 10.8 | 43.2 | 67.6 | - | **78.8** |
| Chinese | CLUEWSC (EM) | 85.4 | 87.9 | 90.9 | 89.9 | - | **92.8** |
| | C-Eval (EM) | 76.7 | 76.0 | 86.5 | 68.9 | - | **91.8** |
| | C-SimpleQA (Correct) | 55.4 | 58.7 | **68.0** | 40.3 | - | 63.7 |
</div>
### Distilled Model Evaluation
<div align="center">
| Model | AIME 2024 pass@1 | AIME 2024 cons@64 | MATH-500 pass@1 | GPQA Diamond pass@1 | LiveCodeBench pass@1 | CodeForces rating |
|------------------------------------------|------------------|-------------------|-----------------|----------------------|----------------------|-------------------|
| GPT-4o-0513 | 9.3 | 13.4 | 74.6 | 49.9 | 32.9 | 759 |
| Claude-3.5-Sonnet-1022 | 16.0 | 26.7 | 78.3 | 65.0 | 38.9 | 717 |
| o1-mini | 63.6 | 80.0 | 90.0 | 60.0 | 53.8 | **1820** |
| QwQ-32B-Preview | 44.0 | 60.0 | 90.6 | 54.5 | 41.9 | 1316 |
| DeepSeek-R1-Distill-Qwen-1.5B | 28.9 | 52.7 | 83.9 | 33.8 | 16.9 | 954 |
| DeepSeek-R1-Distill-Qwen-7B | 55.5 | 83.3 | 92.8 | 49.1 | 37.6 | 1189 |
| DeepSeek-R1-Distill-Qwen-14B | 69.7 | 80.0 | 93.9 | 59.1 | 53.1 | 1481 |
| DeepSeek-R1-Distill-Qwen-32B | **72.6** | 83.3 | 94.3 | 62.1 | 57.2 | 1691 |
| DeepSeek-R1-Distill-Llama-8B | 50.4 | 80.0 | 89.1 | 49.0 | 39.6 | 1205 |
| DeepSeek-R1-Distill-Llama-70B | 70.0 | **86.7** | **94.5** | **65.2** | **57.5** | 1633 |
</div>
## 5. Chat Website & API Platform
You can chat with DeepSeek-R1 on DeepSeek's official website: [chat.deepseek.com](https://chat.deepseek.com), and switch on the button "DeepThink"
We also provide OpenAI-Compatible API at DeepSeek Platform: [platform.deepseek.com](https://platform.deepseek.com/)
## 6. How to Run Locally
### DeepSeek-R1 Models
Please visit [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repo for more information about running DeepSeek-R1 locally.
**NOTE: Hugging Face's Transformers has not been directly supported yet.**
### DeepSeek-R1-Distill Models
DeepSeek-R1-Distill models can be utilized in the same manner as Qwen or Llama models.
For instance, you can easily start a service using [vLLM](https://github.com/vllm-project/vllm):
```shell
vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --tensor-parallel-size 2 --max-model-len 32768 --enforce-eager
```
You can also easily start a service using [SGLang](https://github.com/sgl-project/sglang)
```bash
python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --trust-remote-code --tp 2
```
### Usage Recommendations
**We recommend adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:**
1. Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
2. **Avoid adding a system prompt; all instructions should be contained within the user prompt.**
3. For mathematical problems, it is advisable to include a directive in your prompt such as: "Please reason step by step, and put your final answer within \boxed{}."
4. When evaluating model performance, it is recommended to conduct multiple tests and average the results.
Additionally, we have observed that the DeepSeek-R1 series models tend to bypass thinking pattern (i.e., outputting "\<think\>\n\n\</think\>") when responding to certain queries, which can adversely affect the model's performance.
**To ensure that the model engages in thorough reasoning, we recommend enforcing the model to initiate its response with "\<think\>\n" at the beginning of every output.**
## 7. License
This code repository and the model weights are licensed under the [MIT License](https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE).
DeepSeek-R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs. Please note that:
- DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, DeepSeek-R1-Distill-Qwen-14B and DeepSeek-R1-Distill-Qwen-32B are derived from [Qwen-2.5 series](https://github.com/QwenLM/Qwen2.5), which are originally licensed under [Apache 2.0 License](https://huggingface.co/Qwen/Qwen2.5-1.5B/blob/main/LICENSE), and now finetuned with 800k samples curated with DeepSeek-R1.
- DeepSeek-R1-Distill-Llama-8B is derived from Llama3.1-8B-Base and is originally licensed under [llama3.1 license](https://huggingface.co/meta-llama/Llama-3.1-8B/blob/main/LICENSE).
- DeepSeek-R1-Distill-Llama-70B is derived from Llama3.3-70B-Instruct and is originally licensed under [llama3.3 license](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE).
## 8. Citation
```
@misc{deepseekai2025deepseekr1incentivizingreasoningcapability,
title={DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning},
author={DeepSeek-AI},
year={2025},
eprint={2501.12948},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2501.12948},
}
```
## 9. Contact
If you have any questions, please raise an issue or contact us at [[email protected]]([email protected]).
|
Mungert/OLMo-2-0325-32B-Instruct-GGUF | Mungert | "2025-05-06T22:56:28Z" | 3,488 | 2 | transformers | [
"transformers",
"gguf",
"text-generation",
"en",
"dataset:allenai/RLVR-GSM-MATH-IF-Mixed-Constraints",
"arxiv:2501.00656",
"arxiv:2411.15124",
"base_model:allenai/OLMo-2-0325-32B-DPO",
"base_model:quantized:allenai/OLMo-2-0325-32B-DPO",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-02T05:04:04Z" | ---
license: apache-2.0
language:
- en
datasets:
- allenai/RLVR-GSM-MATH-IF-Mixed-Constraints
base_model:
- allenai/OLMo-2-0325-32B-DPO
pipeline_tag: text-generation
library_name: transformers
---
# <span style="color: #7FFF7F;">OLMo-2-0325-32B-Instruct GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `OLMo-2-0325-32B-Instruct-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `OLMo-2-0325-32B-Instruct-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `OLMo-2-0325-32B-Instruct-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `OLMo-2-0325-32B-Instruct-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `OLMo-2-0325-32B-Instruct-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `OLMo-2-0325-32B-Instruct-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `OLMo-2-0325-32B-Instruct-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `OLMo-2-0325-32B-Instruct-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `OLMo-2-0325-32B-Instruct-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `OLMo-2-0325-32B-Instruct-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `OLMo-2-0325-32B-Instruct-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
<img alt="OLMo Logo" src="https://huggingface.co/datasets/allenai/blog-images/resolve/main/olmo2/olmo.png" width="242px">
OLMo 2 32B Instruct March 2025 is post-trained variant of the [OLMo-2 32B March 2025](https://huggingface.co/allenai/OLMo-2-0325-32B/) model, which has undergone supervised finetuning on an OLMo-specific variant of the [Tülu 3 dataset](https://huggingface.co/datasets/allenai/tulu-3-sft-olmo-2-mixture), further DPO training on [this dataset](https://huggingface.co/datasets/allenai/olmo-2-0325-32b-preference-mix), and final RLVR training on [this dataset](https://huggingface.co/datasets/allenai/RLVR-GSM-MATH-IF-Mixed-Constraints).
Tülu 3 is designed for state-of-the-art performance on a diversity of tasks in addition to chat, such as MATH, GSM8K, and IFEval.
Check out the [OLMo 2 paper](https://arxiv.org/abs/2501.00656) or [Tülu 3 paper](https://arxiv.org/abs/2411.15124) for more details!
OLMo is a series of **O**pen **L**anguage **Mo**dels designed to enable the science of language models.
These models are trained on the Dolma dataset. We are releasing all code, checkpoints, logs, and associated training details.
## Model description
- **Model type:** A model trained on a mix of publicly available, synthetic and human-created datasets.
- **Language(s) (NLP):** Primarily English
- **License:** Apache 2.0
- **Finetuned from model:** allenai/OLMo-2-0325-32B-DPO
### Model Sources
- **Project Page:** https://allenai.org/olmo
- **Repositories:**
- Core repo (training, inference, fine-tuning etc.): https://github.com/allenai/OLMo-core
- Evaluation code: https://github.com/allenai/olmes
- Further fine-tuning code: https://github.com/allenai/open-instruct
- **Paper:** https://arxiv.org/abs/2501.00656
- **Demo:** https://playground.allenai.org/
## Installation
OLMo 2 will be supported in the next version of Transformers, and you need to install it from the main branch using:
```bash
pip install --upgrade git+https://github.com/huggingface/transformers.git
```
## Using the model
### Loading with HuggingFace
To load the model with HuggingFace, use the following snippet:
```
from transformers import AutoModelForCausalLM
olmo_model = AutoModelForCausalLM.from_pretrained("allenai/OLMo-2-0325-32B-Instruct")
```
### Chat template
*NOTE: This is different than previous OLMo 2 and Tülu 3 models due to a minor change in configuration. It does NOT have the bos token before the rest. Our other models have <|endoftext|> at the beginning of the chat template.*
The chat template for our models is formatted as:
```
<|user|>\nHow are you doing?\n<|assistant|>\nI'm just a computer program, so I don't have feelings, but I'm functioning as expected. How can I assist you today?<|endoftext|>
```
Or with new lines expanded:
```
<|user|>
How are you doing?
<|assistant|>
I'm just a computer program, so I don't have feelings, but I'm functioning as expected. How can I assist you today?<|endoftext|>
```
It is embedded within the tokenizer as well, for `tokenizer.apply_chat_template`.
### System prompt
In Ai2 demos, we use this system prompt by default:
```
You are OLMo 2, a helpful and harmless AI Assistant built by the Allen Institute for AI.
```
The model has not been trained with a specific system prompt in mind.
### Intermediate Checkpoints
To facilitate research on RL finetuning, we have released our intermediate checkpoints during the model's RLVR training.
The model weights are saved every 20 training steps, and can be accessible in the revisions of the HuggingFace repository.
For example, you can load with:
```
olmo_model = AutoModelForCausalLM.from_pretrained("allenai/OLMo-2-0325-32B-Instruct", revision="step_200")
```
### Bias, Risks, and Limitations
The OLMo-2 models have limited safety training, but are not deployed automatically with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so).
See the Falcon 180B model card for an example of this.
## Performance
| Model | Average | AlpacaEval 2 LC | BBH | DROP | GSM8k | IFEval | MATH | MMLU | Safety | PopQA | TruthQA |
|-------|---------|------|-----|------|-------|--------|------|------|--------|-------|---------|
| **Closed API models** | | | | | | | | | | | |
| GPT-3.5 Turbo 0125 | 59.6 | 38.7 | 66.6 | 70.2 | 74.3 | 66.9 | 41.2 | 70.2 | 69.1 | 45.0 | 62.9 |
| GPT 4o Mini 2024-07-18 | 65.7 | 49.7 | 65.9 | 36.3 | 83.0 | 83.5 | 67.9 | 82.2 | 84.9 | 39.0 | 64.8 |
| **Open weights models** | | | | | | | | | | | |
| Mistral-Nemo-Instruct-2407 | 50.9 | 45.8 | 54.6 | 23.6 | 81.4 | 64.5 | 31.9 | 70.0 | 52.7 | 26.9 | 57.7 |
| Ministral-8B-Instruct | 52.1 | 31.4 | 56.2 | 56.2 | 80.0 | 56.4 | 40.0 | 68.5 | 56.2 | 20.2 | 55.5 |
| Gemma-2-27b-it | 61.3 | 49.0 | 72.7 | 67.5 | 80.7 | 63.2 | 35.1 | 70.7 | 75.9 | 33.9 | 64.6 |
| Qwen2.5-32B | 66.5 | 39.1 | 82.3 | 48.3 | 87.5 | 82.4 | 77.9 | 84.7 | 82.4 | 26.1 | 70.6 |
| Mistral-Small-24B | 67.6 | 43.2 | 80.1 | 78.5 | 87.2 | 77.3 | 65.9 | 83.7 | 66.5 | 24.4 | 68.1 |
| Llama-3.1-70B | 70.0 | 32.9 | 83.0 | 77.0 | 94.5 | 88.0 | 56.2 | 85.2 | 76.4 | 46.5 | 66.8 |
| Llama-3.3-70B | 73.0 | 36.5 | 85.8 | 78.0 | 93.6 | 90.8 | 71.8 | 85.9 | 70.4 | 48.2 | 66.1 |
| Gemma-3-27b-it | - | 63.4 | 83.7 | 69.2 | 91.1 | - | - | 81.8 | - | 30.9 | - |
| **Fully open models** | | | | | | | | | | | |
| OLMo-2-7B-1124-Instruct | 55.7 | 31.0 | 48.5 | 58.9 | 85.2 | 75.6 | 31.3 | 63.9 | 81.2 | 24.6 | 56.3 |
| OLMo-2-13B-1124-Instruct | 61.4 | 37.5 | 58.4 | 72.1 | 87.4 | 80.4 | 39.7 | 68.6 | 77.5 | 28.8 | 63.9 |
| **OLMo-2-32B-0325-SFT** | 61.7 | 16.9 | 69.7 | 77.2 | 78.4 | 72.4 | 35.9 | 76.1 | 93.8 | 35.4 | 61.3 |
| **OLMo-2-32B-0325-DPO** | 68.8 | 44.1 | 70.2 | 77.5 | 85.7 | 83.8 | 46.8 | 78.0 | 91.9 | 36.4 | 73.5 |
| **OLMo-2-32B-0325-Instruct** | 68.8 | 42.8 | 70.6 | 78.0 | 87.6 | 85.6 | 49.7 | 77.3 | 85.9 | 37.5 | 73.2 |
## Learning curves
Below is the training curves for `allenai/OLMo-2-0325-32B-Instruct`. The model was trained using 5 8xH100 nodes.
Below are the core eval scores over steps for `allenai/OLMo-2-0325-32B-Instruct` (note we took step `320` as the final checkpoint, corresponding to episode `573,440`):
Below are the other eval scores over steps for `allenai/OLMo-2-0325-32B-Instruct`:
## Reproduction command
The command below is copied directly from the tracked training job:
```bash
# clone and check out commit
git clone https://github.com/allenai/open-instruct.git
# this should be the correct commit, the main thing is to have the vllm monkey patch for
# 32b olmo https://github.com/allenai/open-instruct/blob/894ffa236319bc6c26c346240a7e4ee04ba0bd31/open_instruct/vllm_utils2.py#L37-L59
git checkout a51dc98525eec01de6e8a24c071f42dce407d738
uv sync
uv sync --extra compile
# note that you may need 5 8xH100 nodes for the training.
# so please setup ray properly, e.g., https://github.com/allenai/open-instruct/blob/main/docs/tulu3.md#llama-31-tulu-3-70b-reproduction
python open_instruct/grpo_vllm_thread_ray_gtrl.py \
--exp_name 0310_olmo2_32b_grpo_12818 \
--beta 0.01 \
--local_mini_batch_size 32 \
--number_samples_per_prompt 16 \
--output_dir output \
--local_rollout_batch_size 4 \
--kl_estimator kl3 \
--learning_rate 5e-7 \
--dataset_mixer_list allenai/RLVR-GSM-MATH-IF-Mixed-Constraints 1.0 \
--dataset_mixer_list_splits train \
--dataset_mixer_eval_list allenai/RLVR-GSM-MATH-IF-Mixed-Constraints 16 \
--dataset_mixer_eval_list_splits train \
--max_token_length 2048 \
--max_prompt_token_length 2048 \
--response_length 2048 \
--model_name_or_path allenai/OLMo-2-0325-32B-DPO \
--non_stop_penalty \
--stop_token eos \
--temperature 1.0 \
--ground_truths_key ground_truth \
--chat_template_name tulu \
--sft_messages_key messages \
--eval_max_length 4096 \
--total_episodes 10000000 \
--penalty_reward_value 0.0 \
--deepspeed_stage 3 \
--no_gather_whole_model \
--per_device_train_batch_size 2 \
--local_rollout_forward_batch_size 2 \
--actor_num_gpus_per_node 8 8 8 4 \
--num_epochs 1 \
--vllm_tensor_parallel_size 1 \
--vllm_num_engines 12 \
--lr_scheduler_type constant \
--apply_verifiable_reward true \
--seed 1 \
--num_evals 30 \
--save_freq 20 \
--reward_model_multiplier 0.0 \
--no_try_launch_beaker_eval_jobs \
--try_launch_beaker_eval_jobs_on_weka \
--gradient_checkpointing \
--with_tracking
```
## License and use
OLMo 2 is licensed under the Apache 2.0 license.
OLMo 2 is intended for research and educational use.
For more information, please see our [Responsible Use Guidelines](https://allenai.org/responsible-use).
This model has been fine-tuned using a dataset mix with outputs generated from third party models and are subject to additional terms: [Gemma Terms of Use](https://ai.google.dev/gemma/terms).
## Citation
```bibtex
@article{olmo20242olmo2furious,
title={2 OLMo 2 Furious},
author={Team OLMo and Pete Walsh and Luca Soldaini and Dirk Groeneveld and Kyle Lo and Shane Arora and Akshita Bhagia and Yuling Gu and Shengyi Huang and Matt Jordan and Nathan Lambert and Dustin Schwenk and Oyvind Tafjord and Taira Anderson and David Atkinson and Faeze Brahman and Christopher Clark and Pradeep Dasigi and Nouha Dziri and Michal Guerquin and Hamish Ivison and Pang Wei Koh and Jiacheng Liu and Saumya Malik and William Merrill and Lester James V. Miranda and Jacob Morrison and Tyler Murray and Crystal Nam and Valentina Pyatkin and Aman Rangapur and Michael Schmitz and Sam Skjonsberg and David Wadden and Christopher Wilhelm and Michael Wilson and Luke Zettlemoyer and Ali Farhadi and Noah A. Smith and Hannaneh Hajishirzi},
year={2024},
eprint={2501.00656},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2501.00656},
}
``` |
gavrilstep/ea157c8a-5a79-4fd5-b389-84738be97d6f | gavrilstep | "2025-05-06T22:56:10Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"llama",
"axolotl",
"generated_from_trainer",
"base_model:elyza/Llama-3-ELYZA-JP-8B",
"base_model:adapter:elyza/Llama-3-ELYZA-JP-8B",
"license:llama3",
"8-bit",
"bitsandbytes",
"region:us"
] | null | "2025-05-06T22:33:31Z" | ---
library_name: peft
license: llama3
base_model: elyza/Llama-3-ELYZA-JP-8B
tags:
- axolotl
- generated_from_trainer
model-index:
- name: ea157c8a-5a79-4fd5-b389-84738be97d6f
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.4.1`
```yaml
absolute_data_files: false
adapter: lora
base_model: elyza/Llama-3-ELYZA-JP-8B
bf16: true
chat_template: llama3
dataset_prepared_path: /workspace/axolotl
datasets:
- data_files:
- dd789f5e7fb257d6_train_data.json
ds_type: json
format: custom
path: /workspace/input_data/dd789f5e7fb257d6_train_data.json
type:
field_input: reasoning (reasoning_content)
field_instruction: question
field_output: response (content)
format: '{instruction} {input}'
no_input_format: '{instruction}'
system_format: '{system}'
system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 1
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_clipping: 0.55
group_by_length: false
hub_model_id: gavrilstep/ea157c8a-5a79-4fd5-b389-84738be97d6f
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 1.0e-06
load_in_4bit: true
load_in_8bit: true
local_rank: null
logging_steps: 1
lora_alpha: 96
lora_dropout: 0.01
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 48
lora_target_linear: true
lr_scheduler: cosine
max_steps: 150
micro_batch_size: 4
mixed_precision: bf16
mlflow_experiment_name: /tmp/dd789f5e7fb257d6_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 1
sequence_len: 2048
special_tokens:
pad_token: <|eot_id|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: b8f59716-406a-400a-ac23-4cce40b4ac8a
wandb_project: s56-7
wandb_run: your_name
wandb_runid: b8f59716-406a-400a-ac23-4cce40b4ac8a
warmup_steps: 5
weight_decay: 0.01
xformers_attention: true
```
</details><br>
# ea157c8a-5a79-4fd5-b389-84738be97d6f
This model is a fine-tuned version of [elyza/Llama-3-ELYZA-JP-8B](https://huggingface.co/elyza/Llama-3-ELYZA-JP-8B) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 1.3065
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-06
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 5
- training_steps: 150
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 1.2066 | 0.0301 | 150 | 1.3065 |
### Framework versions
- PEFT 0.13.2
- Transformers 4.46.0
- Pytorch 2.5.0+cu124
- Datasets 3.0.1
- Tokenizers 0.20.1 |
Mungert/Llama-3.1-Nemotron-Nano-8B-v1-GGUF | Mungert | "2025-05-06T22:56:05Z" | 8,998 | 5 | transformers | [
"transformers",
"gguf",
"nvidia",
"llama-3",
"pytorch",
"text-generation",
"en",
"arxiv:2502.00203",
"license:other",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-03-21T19:44:49Z" | ---
library_name: transformers
license: other
license_name: nvidia-open-model-license
license_link: >-
https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
pipeline_tag: text-generation
language:
- en
tags:
- nvidia
- llama-3
- pytorch
---
# <span style="color: #7FFF7F;">Llama-3.1-Nemotron-Nano-8B-v1 GGUF Models</span>
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Llama-3.1-Nemotron-Nano-8B-v1-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Llama-3.1-Nemotron-Nano-8B-v1-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Llama-3.1-Nemotron-Nano-8B-v1-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Llama-3.1-Nemotron-Nano-8B-v1-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Llama-3.1-Nemotron-Nano-8B-v1-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Llama-3.1-Nemotron-Nano-8B-v1-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Llama-3.1-Nemotron-Nano-8B-v1-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Llama-3.1-Nemotron-Nano-8B-v1-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Llama-3.1-Nemotron-Nano-8B-v1
## Model Overview
Llama-3.1-Nemotron-Nano-8B-v1 is a large language model (LLM) which is a derivative of [Meta Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) (AKA the reference model). It is a reasoning model that is post trained for reasoning, human chat preferences, and tasks, such as RAG and tool calling.
Llama-3.1-Nemotron-Nano-8B-v1 is a model which offers a great tradeoff between model accuracy and efficiency. It is created from Llama 3.1 8B Instruct and offers improvements in model accuracy. The model fits on a single RTX GPU and can be used locally. The model supports a context length of 128K.
This model underwent a multi-phase post-training process to enhance both its reasoning and non-reasoning capabilities. This includes a supervised fine-tuning stage for Math, Code, Reasoning, and Tool Calling as well as multiple reinforcement learning (RL) stages using REINFORCE (RLOO) and Online Reward-aware Preference Optimization (RPO) algorithms for both chat and instruction-following. The final model checkpoint is obtained after merging the final SFT and Online RPO checkpoints. Improved using Qwen.
This model is part of the Llama Nemotron Collection. You can find the other model(s) in this family here:
[Llama-3.3-Nemotron-Super-49B-v1](https://huggingface.co/nvidia/Llama-3.3-Nemotron-Super-49B-v1)
This model is ready for commercial use.
## License/Terms of Use
GOVERNING TERMS: Your use of this model is governed by the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). Additional Information: [Llama 3.1 Community License Agreement](https://www.llama.com/llama3_1/license/). Built with Llama.
**Model Developer:** NVIDIA
**Model Dates:** Trained between August 2024 and March 2025
**Data Freshness:** The pretraining data has a cutoff of 2023 per Meta Llama 3.1 8B
## Use Case:
Developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks. Balance of model accuracy and compute efficiency (the model fits on a single RTX GPU and can be used locally).
## Release Date: <br>
3/18/2025 <br>
## References
- [\[2502.00203\] Reward-aware Preference Optimization: A Unified Mathematical Framework for Model Alignment](https://arxiv.org/abs/2502.00203)
## Model Architecture
**Architecture Type:** Dense decoder-only Transformer model
**Network Architecture:** Llama 3.1 8B Instruct
## Intended use
Llama-3.1-Nemotron-Nano-8B-v1 is a general purpose reasoning and chat model intended to be used in English and coding languages. Other non-English languages (German, French, Italian, Portuguese, Hindi, Spanish, and Thai) are also supported.
# Input:
- **Input Type:** Text
- **Input Format:** String
- **Input Parameters:** One-Dimensional (1D)
- **Other Properties Related to Input:** Context length up to 131,072 tokens
## Output:
- **Output Type:** Text
- **Output Format:** String
- **Output Parameters:** One-Dimensional (1D)
- **Other Properties Related to Output:** Context length up to 131,072 tokens
## Model Version:
1.0 (3/18/2025)
## Software Integration
- **Runtime Engine:** NeMo 24.12 <br>
- **Recommended Hardware Microarchitecture Compatibility:**
- NVIDIA Hopper
- NVIDIA Ampere
## Quick Start and Usage Recommendations:
1. Reasoning mode (ON/OFF) is controlled via the system prompt, which must be set as shown in the example below. All instructions should be contained within the user prompt
2. We recommend setting temperature to `0.6`, and Top P to `0.95` for Reasoning ON mode
3. We recommend using greedy decoding for Reasoning OFF mode
4. We have provided a list of prompts to use for evaluation for each benchmark where a specific template is required
You can try this model out through the preview API, using this link: [Llama-3.1-Nemotron-Nano-8B-v1](https://build.nvidia.com/nvidia/llama-3_1-nemotron-nano-8b-v1).
See the snippet below for usage with Hugging Face Transformers library. Reasoning mode (ON/OFF) is controlled via system prompt. Please see the example below.
Our code requires the transformers package version to be `4.44.2` or higher.
### Example of “Reasoning On:”
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
temperature=0.6,
top_p=0.95,
**model_kwargs
)
# Thinking can be "on" or "off"
thinking = "on"
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}]))
```
### Example of “Reasoning Off:”
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
do_sample=False,
**model_kwargs
)
# Thinking can be "on" or "off"
thinking = "off"
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}]))
```
For some prompts, even though thinking is disabled, the model emergently prefers to think before responding. But if desired, the users can prevent it by pre-filling the assistant response.
```python
import torch
import transformers
model_id = "nvidia/Llama-3.1-Nemotron-Nano-8B-v1"
model_kwargs = {"torch_dtype": torch.bfloat16, "device_map": "auto"}
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id
# Thinking can be "on" or "off"
thinking = "off"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
tokenizer=tokenizer,
max_new_tokens=32768,
do_sample=False,
**model_kwargs
)
print(pipeline([{"role": "system", "content": f"detailed thinking {thinking}"}, {"role": "user", "content": "Solve x*(sin(x)+2)=0"}, {"role":"assistant", "content":"<think>\n</think>"}]))
```
## Inference:
**Engine:** Transformers
**Test Hardware:**
- BF16:
- 1x RTX 50 Series GPUs
- 1x RTX 40 Series GPUs
- 1x RTX 30 Series GPUs
- 1x H100-80GB GPU
- 1x A100-80GB GPU
**Preferred/Supported] Operating System(s):** Linux <br>
## Training Datasets
A large variety of training data was used for the post-training pipeline, including manually annotated data and synthetic data.
The data for the multi-stage post-training phases for improvements in Code, Math, and Reasoning is a compilation of SFT and RL data that supports improvements of math, code, general reasoning, and instruction following capabilities of the original Llama instruct model.
Prompts have been sourced from either public and open corpus or synthetically generated. Responses were synthetically generated by a variety of models, with some prompts containing responses for both Reasoning On and Off modes, to train the model to distinguish between two modes.
**Data Collection for Training Datasets:** <br>
* Hybrid: Automated, Human, Synthetic <br>
**Data Labeling for Training Datasets:** <br>
* N/A <br>
## Evaluation Datasets
We used the datasets listed below to evaluate Llama-3.1-Nemotron-Nano-8B-v1.
**Data Collection for Evaluation Datasets:** Hybrid: Human/Synthetic
**Data Labeling for Evaluation Datasets:** Hybrid: Human/Synthetic/Automatic
## Evaluation Results
These results contain both “Reasoning On”, and “Reasoning Off”. We recommend using temperature=`0.6`, top_p=`0.95` for “Reasoning On” mode, and greedy decoding for “Reasoning Off” mode. All evaluations are done with 32k sequence length. We run the benchmarks up to 16 times and average the scores to be more accurate.
> NOTE: Where applicable, a Prompt Template will be provided. While completing benchmarks, please ensure that you are parsing for the correct output format as per the provided prompt in order to reproduce the benchmarks seen below.
### MT-Bench
| Reasoning Mode | Score |
|--------------|------------|
| Reasoning Off | 7.9 |
| Reasoning On | 8.1 |
### MATH500
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 36.6% |
| Reasoning On | 95.4% |
User Prompt Template:
```
"Below is a math question. I want you to reason through the steps and then give a final answer. Your final answer should be in \boxed{}.\nQuestion: {question}"
```
### AIME25
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 0% |
| Reasoning On | 47.1% |
User Prompt Template:
```
"Below is a math question. I want you to reason through the steps and then give a final answer. Your final answer should be in \boxed{}.\nQuestion: {question}"
```
### GPQA-D
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 39.4% |
| Reasoning On | 54.1% |
User Prompt Template:
```
"What is the correct answer to this question: {question}\nChoices:\nA. {option_A}\nB. {option_B}\nC. {option_C}\nD. {option_D}\nLet's think step by step, and put the final answer (should be a single letter A, B, C, or D) into a \boxed{}"
```
### IFEval Average
| Reasoning Mode | Strict:Prompt | Strict:Instruction |
|--------------|------------|------------|
| Reasoning Off | 74.7% | 82.1% |
| Reasoning On | 71.9% | 79.3% |
### BFCL v2 Live
| Reasoning Mode | Score |
|--------------|------------|
| Reasoning Off | 63.9% |
| Reasoning On | 63.6% |
User Prompt Template:
```
<AVAILABLE_TOOLS>{functions}</AVAILABLE_TOOLS>
{user_prompt}
```
### MBPP 0-shot
| Reasoning Mode | pass@1 |
|--------------|------------|
| Reasoning Off | 66.1% |
| Reasoning On | 84.6% |
User Prompt Template:
````
You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.
@@ Instruction
Here is the given problem and test examples:
{prompt}
Please use the python programming language to solve this problem.
Please make sure that your code includes the functions from the test samples and that the input and output formats of these functions match the test samples.
Please return all completed codes in one code block.
This code block should be in the following format:
```python
# Your codes here
```
````
## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ [Explainability](explainability.md), [Bias](bias.md), [Safety & Security](safety.md), and [Privacy](privacy.md) Subcards.
Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/). |
vermoney/f2b291fa-5267-4c0c-9a05-d5e920f9fa72 | vermoney | "2025-05-06T22:55:12Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"llama",
"axolotl",
"generated_from_trainer",
"base_model:elyza/Llama-3-ELYZA-JP-8B",
"base_model:adapter:elyza/Llama-3-ELYZA-JP-8B",
"license:llama3",
"4-bit",
"bitsandbytes",
"region:us"
] | null | "2025-05-06T22:37:40Z" | ---
library_name: peft
license: llama3
base_model: elyza/Llama-3-ELYZA-JP-8B
tags:
- axolotl
- generated_from_trainer
model-index:
- name: f2b291fa-5267-4c0c-9a05-d5e920f9fa72
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.4.1`
```yaml
adapter: lora
base_model: elyza/Llama-3-ELYZA-JP-8B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
- dd789f5e7fb257d6_train_data.json
ds_type: json
format: custom
path: /workspace/input_data/dd789f5e7fb257d6_train_data.json
type:
field_input: reasoning (reasoning_content)
field_instruction: question
field_output: response (content)
format: '{instruction} {input}'
no_input_format: '{instruction}'
system_format: '{system}'
system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 1
flash_attention: true
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_clipping: 0.5
group_by_length: false
hub_model_id: vermoney/f2b291fa-5267-4c0c-9a05-d5e920f9fa72
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 5.0e-06
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
max_steps: 400
micro_batch_size: 8
mixed_precision: bf16
mlflow_experiment_name: /tmp/dd789f5e7fb257d6_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 1
sequence_len: 1024
special_tokens:
pad_token: <|eot_id|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: b8f59716-406a-400a-ac23-4cce40b4ac8a
wandb_project: s56-9
wandb_run: your_name
wandb_runid: b8f59716-406a-400a-ac23-4cce40b4ac8a
warmup_steps: 20
weight_decay: 0.01
xformers_attention: true
```
</details><br>
# f2b291fa-5267-4c0c-9a05-d5e920f9fa72
This model is a fine-tuned version of [elyza/Llama-3-ELYZA-JP-8B](https://huggingface.co/elyza/Llama-3-ELYZA-JP-8B) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.9414
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 20
- training_steps: 400
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.9823 | 0.1604 | 400 | 0.9414 |
### Framework versions
- PEFT 0.13.2
- Transformers 4.46.0
- Pytorch 2.5.0+cu124
- Datasets 3.0.1
- Tokenizers 0.20.1 |
Mungert/Qwen3-8B-GGUF | Mungert | "2025-05-06T22:55:11Z" | 1,269 | 4 | transformers | [
"transformers",
"gguf",
"text-generation",
"arxiv:2309.00071",
"base_model:Qwen/Qwen3-8B-Base",
"base_model:quantized:Qwen/Qwen3-8B-Base",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | text-generation | "2025-04-30T06:20:32Z" | ---
library_name: transformers
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE
pipeline_tag: text-generation
base_model:
- Qwen/Qwen3-8B-Base
---
# <span style="color: #7FFF7F;">Qwen3-8B GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Qwen3-8B-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Qwen3-8B-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Qwen3-8B-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Qwen3-8B-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Qwen3-8B-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Qwen3-8B-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Qwen3-8B-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Qwen3-8B-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Qwen3-8B-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Qwen3-8B-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Qwen3-8B-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
# Qwen3-8B
<a href="https://chat.qwen.ai/" target="_blank" style="margin: 2px;">
<img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
</a>
## Qwen3 Highlights
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features:
- **Uniquely support of seamless switching between thinking mode** (for complex logical reasoning, math, and coding) and **non-thinking mode** (for efficient, general-purpose dialogue) **within single model**, ensuring optimal performance across various scenarios.
- **Significantly enhancement in its reasoning capabilities**, surpassing previous QwQ (in thinking mode) and Qwen2.5 instruct models (in non-thinking mode) on mathematics, code generation, and commonsense logical reasoning.
- **Superior human preference alignment**, excelling in creative writing, role-playing, multi-turn dialogues, and instruction following, to deliver a more natural, engaging, and immersive conversational experience.
- **Expertise in agent capabilities**, enabling precise integration with external tools in both thinking and unthinking modes and achieving leading performance among open-source models in complex agent-based tasks.
- **Support of 100+ languages and dialects** with strong capabilities for **multilingual instruction following** and **translation**.
## Model Overview
**Qwen3-8B** has the following features:
- Type: Causal Language Models
- Training Stage: Pretraining & Post-training
- Number of Parameters: 8.2B
- Number of Paramaters (Non-Embedding): 6.95B
- Number of Layers: 36
- Number of Attention Heads (GQA): 32 for Q and 8 for KV
- Context Length: 32,768 natively and [131,072 tokens with YaRN](#processing-long-texts).
For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our [blog](https://qwenlm.github.io/blog/qwen3/), [GitHub](https://github.com/QwenLM/Qwen3), and [Documentation](https://qwen.readthedocs.io/en/latest/).
## Quickstart
The code of Qwen3 has been in the latest Hugging Face `transformers` and we advise you to use the latest version of `transformers`.
With `transformers<4.51.0`, you will encounter the following error:
```
KeyError: 'qwen3'
```
The following contains a code snippet illustrating how to use the model generate content based on given inputs.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Qwen/Qwen3-8B"
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
# conduct text completion
generated_ids = model.generate(
**model_inputs,
max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
# parsing thinking content
try:
# rindex finding 151668 (</think>)
index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
index = 0
thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
print("thinking content:", thinking_content)
print("content:", content)
```
For deployment, you can use `sglang>=0.4.6.post1` or `vllm>=0.8.5` or to create an OpenAI-compatible API endpoint:
- SGLang:
```shell
python -m sglang.launch_server --model-path Qwen/Qwen3-8B --reasoning-parser qwen3
```
- vLLM:
```shell
vllm serve Qwen/Qwen3-8B --enable-reasoning --reasoning-parser deepseek_r1
```
For local use, applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers have also supported Qwen3.
## Switching Between Thinking and Non-Thinking Mode
> [!TIP]
> The `enable_thinking` switch is also available in APIs created by SGLang and vLLM.
> Please refer to our documentation for [SGLang](https://qwen.readthedocs.io/en/latest/deployment/sglang.html#thinking-non-thinking-modes) and [vLLM](https://qwen.readthedocs.io/en/latest/deployment/vllm.html#thinking-non-thinking-modes) users.
### `enable_thinking=True`
By default, Qwen3 has thinking capabilities enabled, similar to QwQ-32B. This means the model will use its reasoning abilities to enhance the quality of generated responses. For example, when explicitly setting `enable_thinking=True` or leaving it as the default value in `tokenizer.apply_chat_template`, the model will engage its thinking mode.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # True is the default value for enable_thinking
)
```
In this mode, the model will generate think content wrapped in a `<think>...</think>` block, followed by the final response.
> [!NOTE]
> For thinking mode, use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0` (the default setting in `generation_config.json`). **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### `enable_thinking=False`
We provide a hard switch to strictly disable the model's thinking behavior, aligning its functionality with the previous Qwen2.5-Instruct models. This mode is particularly useful in scenarios where disabling thinking is essential for enhancing efficiency.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False # Setting enable_thinking=False disables thinking mode
)
```
In this mode, the model will not generate any think content and will not include a `<think>...</think>` block.
> [!NOTE]
> For non-thinking mode, we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### Advanced Usage: Switching Between Thinking and Non-Thinking Modes via User Input
We provide a soft switch mechanism that allows users to dynamically control the model's behavior when `enable_thinking=True`. Specifically, you can add `/think` and `/no_think` to user prompts or system messages to switch the model's thinking mode from turn to turn. The model will follow the most recent instruction in multi-turn conversations.
Here is an example of a multi-turn conversation:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
class QwenChatbot:
def __init__(self, model_name="Qwen/Qwen3-8B"):
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
self.model = AutoModelForCausalLM.from_pretrained(model_name)
self.history = []
def generate_response(self, user_input):
messages = self.history + [{"role": "user", "content": user_input}]
text = self.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
inputs = self.tokenizer(text, return_tensors="pt")
response_ids = self.model.generate(**inputs, max_new_tokens=32768)[0][len(inputs.input_ids[0]):].tolist()
response = self.tokenizer.decode(response_ids, skip_special_tokens=True)
# Update history
self.history.append({"role": "user", "content": user_input})
self.history.append({"role": "assistant", "content": response})
return response
# Example Usage
if __name__ == "__main__":
chatbot = QwenChatbot()
# First input (without /think or /no_think tags, thinking mode is enabled by default)
user_input_1 = "How many r's in strawberries?"
print(f"User: {user_input_1}")
response_1 = chatbot.generate_response(user_input_1)
print(f"Bot: {response_1}")
print("----------------------")
# Second input with /no_think
user_input_2 = "Then, how many r's in blueberries? /no_think"
print(f"User: {user_input_2}")
response_2 = chatbot.generate_response(user_input_2)
print(f"Bot: {response_2}")
print("----------------------")
# Third input with /think
user_input_3 = "Really? /think"
print(f"User: {user_input_3}")
response_3 = chatbot.generate_response(user_input_3)
print(f"Bot: {response_3}")
```
> [!NOTE]
> For API compatibility, when `enable_thinking=True`, regardless of whether the user uses `/think` or `/no_think`, the model will always output a block wrapped in `<think>...</think>`. However, the content inside this block may be empty if thinking is disabled.
> When `enable_thinking=False`, the soft switches are not valid. Regardless of any `/think` or `/no_think` tags input by the user, the model will not generate think content and will not include a `<think>...</think>` block.
## Agentic Use
Qwen3 excels in tool calling capabilities. We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to make the best use of agentic ability of Qwen3. Qwen-Agent encapsulates tool-calling templates and tool-calling parsers internally, greatly reducing coding complexity.
To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
```python
from qwen_agent.agents import Assistant
# Define LLM
llm_cfg = {
'model': 'Qwen3-8B',
# Use the endpoint provided by Alibaba Model Studio:
# 'model_type': 'qwen_dashscope',
# 'api_key': os.getenv('DASHSCOPE_API_KEY'),
# Use a custom endpoint compatible with OpenAI API:
'model_server': 'http://localhost:8000/v1', # api_base
'api_key': 'EMPTY',
# Other parameters:
# 'generate_cfg': {
# # Add: When the response content is `<think>this is the thought</think>this is the answer;
# # Do not add: When the response has been separated by reasoning_content and content.
# 'thought_in_content': True,
# },
}
# Define Tools
tools = [
{'mcpServers': { # You can specify the MCP configuration file
'time': {
'command': 'uvx',
'args': ['mcp-server-time', '--local-timezone=Asia/Shanghai']
},
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
},
'code_interpreter', # Built-in tools
]
# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)
# Streaming generation
messages = [{'role': 'user', 'content': 'https://qwenlm.github.io/blog/ Introduce the latest developments of Qwen'}]
for responses in bot.run(messages=messages):
pass
print(responses)
```
## Processing Long Texts
Qwen3 natively supports context lengths of up to 32,768 tokens. For conversations where the total length (including both input and output) significantly exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively. We have validated the model's performance on context lengths of up to 131,072 tokens using the [YaRN](https://arxiv.org/abs/2309.00071) method.
YaRN is currently supported by several inference frameworks, e.g., `transformers` and `llama.cpp` for local use, `vllm` and `sglang` for deployment. In general, there are two approaches to enabling YaRN for supported frameworks:
- Modifying the model files:
In the `config.json` file, add the `rope_scaling` fields:
```json
{
...,
"rope_scaling": {
"rope_type": "yarn",
"factor": 4.0,
"original_max_position_embeddings": 32768
}
}
```
For `llama.cpp`, you need to regenerate the GGUF file after the modification.
- Passing command line arguments:
For `vllm`, you can use
```shell
vllm serve ... --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}' --max-model-len 131072
```
For `sglang`, you can use
```shell
python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling":{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}}'
```
For `llama-server` from `llama.cpp`, you can use
```shell
llama-server ... --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768
```
> [!IMPORTANT]
> If you encounter the following warning
> ```
> Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'original_max_position_embeddings'}
> ```
> please upgrade `transformers>=4.51.0`.
> [!NOTE]
> All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts.**
> We advise adding the `rope_scaling` configuration only when processing long contexts is required.
> It is also recommended to modify the `factor` as needed. For example, if the typical context length for your application is 65,536 tokens, it would be better to set `factor` as 2.0.
> [!NOTE]
> The default `max_position_embeddings` in `config.json` is set to 40,960. This allocation includes reserving 32,768 tokens for outputs and 8,192 tokens for typical prompts, which is sufficient for most scenarios involving short text processing. If the average context length does not exceed 32,768 tokens, we do not recommend enabling YaRN in this scenario, as it may potentially degrade model performance.
> [!TIP]
> The endpoint provided by Alibaba Model Studio supports dynamic YaRN by default and no extra configuration is needed.
## Best Practices
To achieve optimal performance, we recommend the following settings:
1. **Sampling Parameters**:
- For thinking mode (`enable_thinking=True`), use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0`. **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions.
- For non-thinking mode (`enable_thinking=False`), we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`.
- For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 38,912 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
- **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
- **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
4. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
### Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{qwen3,
title = {Qwen3},
url = {https://qwenlm.github.io/blog/qwen3/},
author = {Qwen Team},
month = {April},
year = {2025}
}
``` |
Mungert/Qwen3-8B-abliterated-GGUF | Mungert | "2025-05-06T22:54:36Z" | 2,198 | 6 | transformers | [
"transformers",
"gguf",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-04-30T07:26:51Z" | ---
library_name: transformers
tags: []
---
# <span style="color: #7FFF7F;">Qwen3-8B-abliterated GGUF Models</span>
## <span style="color: #7F7FFF;">Model Generation Details</span>
This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`19e899c`](https://github.com/ggerganov/llama.cpp/commit/19e899ce21a7c9ffcf8bb2b22269a75f6e078f8f).
## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
### **Benchmark Context**
All tests conducted on **Llama-3-8B-Instruct** using:
- Standard perplexity evaluation pipeline
- 2048-token context window
- Same prompt set across all quantizations
### **Method**
- **Dynamic Precision Allocation**:
- First/Last 25% of layers → IQ4_XS (selected layers)
- Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
- **Critical Component Protection**:
- Embeddings/output layers use Q5_K
- Reduces error propagation by 38% vs standard 1-2bit
### **Quantization Performance Comparison (Llama-3-8B)**
| Quantization | Standard PPL | DynamicGate PPL | Δ PPL | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
| IQ2_XXS | 11.30 | 9.84 | -12.9% | 2.5G | 2.6G | +0.1G | 234s | 246s |
| IQ2_XS | 11.72 | 11.63 | -0.8% | 2.7G | 2.8G | +0.1G | 242s | 246s |
| IQ2_S | 14.31 | 9.02 | -36.9% | 2.7G | 2.9G | +0.2G | 238s | 244s |
| IQ1_M | 27.46 | 15.41 | -43.9% | 2.2G | 2.5G | +0.3G | 206s | 212s |
| IQ1_S | 53.07 | 32.00 | -39.7% | 2.1G | 2.4G | +0.3G | 184s | 209s |
**Key**:
- PPL = Perplexity (lower is better)
- Δ PPL = Percentage change from standard to DynamicGate
- Speed = Inference time (CPU avx2, 2048 token context)
- Size differences reflect mixed quantization overhead
**Key Improvements:**
- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
**Tradeoffs:**
- All variants have modest size increases (0.1-0.3GB)
- Inference speeds remain comparable (<5% difference)
### **When to Use These Models**
📌 **Fitting models into GPU VRAM**
✔ **Memory-constrained deployments**
✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated
✔ **Research** into ultra-low-bit quantization
## **Choosing the Right Model Format**
Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.
### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**
- A 16-bit floating-point format designed for **faster computation** while retaining good precision.
- Provides **similar dynamic range** as FP32 but with **lower memory usage**.
- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).
- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.
📌 **Use BF16 if:**
✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).
✔ You want **higher precision** while saving memory.
✔ You plan to **requantize** the model into another format.
📌 **Avoid BF16 if:**
❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.
---
### **F16 (Float 16) – More widely supported than BF16**
- A 16-bit floating-point **high precision** but with less of range of values than BF16.
- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).
- Slightly lower numerical precision than BF16 but generally sufficient for inference.
📌 **Use F16 if:**
✔ Your hardware supports **FP16** but **not BF16**.
✔ You need a **balance between speed, memory usage, and accuracy**.
✔ You are running on a **GPU** or another device optimized for FP16 computations.
📌 **Avoid F16 if:**
❌ Your device lacks **native FP16 support** (it may run slower than expected).
❌ You have memory limitations.
---
### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**
Quantization reduces model size and memory usage while maintaining as much accuracy as possible.
- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.
- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.
📌 **Use Quantized Models if:**
✔ You are running inference on a **CPU** and need an optimized model.
✔ Your device has **low VRAM** and cannot load full-precision models.
✔ You want to reduce **memory footprint** while keeping reasonable accuracy.
📌 **Avoid Quantized Models if:**
❌ You need **maximum accuracy** (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).
---
### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**
These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.
- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.
- **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.
- **Trade-off**: Lower accuracy compared to higher-bit quantizations.
- **IQ3_S**: Small block size for **maximum memory efficiency**.
- **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.
- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.
- **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.
- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.
- **Use case**: Best for **low-memory devices** where **Q6_K** is too large.
- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.
- **Use case**: Best for **ARM-based devices** or **low-memory environments**.
---
### **Summary Table: Model Format Selection**
| Model Format | Precision | Memory Usage | Device Requirements | Best Use Case |
|--------------|------------|---------------|----------------------|---------------|
| **BF16** | Highest | High | BF16-supported GPU/CPUs | High-speed inference with reduced memory |
| **F16** | High | High | FP16-supported devices | GPU inference when BF16 isn't available |
| **Q4_K** | Medium Low | Low | CPU or Low-VRAM devices | Best for memory-constrained environments |
| **Q6_K** | Medium | Moderate | CPU with more memory | Better accuracy while still being quantized |
| **Q8_0** | High | Moderate | CPU or GPU with enough VRAM | Best accuracy among quantized models |
| **IQ3_XS** | Very Low | Very Low | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |
| **Q4_0** | Low | Low | ARM or low-memory devices | llama.cpp can optimize for ARM devices |
---
## **Included Files & Details**
### `Qwen3-8B-abliterated-bf16.gguf`
- Model weights preserved in **BF16**.
- Use this if you want to **requantize** the model into a different format.
- Best if your device supports **BF16 acceleration**.
### `Qwen3-8B-abliterated-f16.gguf`
- Model weights stored in **F16**.
- Use if your device supports **FP16**, especially if BF16 is not available.
### `Qwen3-8B-abliterated-bf16-q8_0.gguf`
- **Output & embeddings** remain in **BF16**.
- All other layers quantized to **Q8_0**.
- Use if your device supports **BF16** and you want a quantized version.
### `Qwen3-8B-abliterated-f16-q8_0.gguf`
- **Output & embeddings** remain in **F16**.
- All other layers quantized to **Q8_0**.
### `Qwen3-8B-abliterated-q4_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q4_K**.
- Good for **CPU inference** with limited memory.
### `Qwen3-8B-abliterated-q4_k_s.gguf`
- Smallest **Q4_K** variant, using less memory at the cost of accuracy.
- Best for **very low-memory setups**.
### `Qwen3-8B-abliterated-q6_k.gguf`
- **Output & embeddings** quantized to **Q8_0**.
- All other layers quantized to **Q6_K** .
### `Qwen3-8B-abliterated-q8_0.gguf`
- Fully **Q8** quantized model for better accuracy.
- Requires **more memory** but offers higher precision.
### `Qwen3-8B-abliterated-iq3_xs.gguf`
- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.
- Best for **ultra-low-memory devices**.
### `Qwen3-8B-abliterated-iq3_m.gguf`
- **IQ3_M** quantization, offering a **medium block size** for better accuracy.
- Suitable for **low-memory devices**.
### `Qwen3-8B-abliterated-q4_0.gguf`
- Pure **Q4_0** quantization, optimized for **ARM devices**.
- Best for **low-memory environments**.
- Prefer IQ4_NL for better accuracy.
# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
❤ **Please click "Like" if you find this useful!**
Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:
👉 [Free Network Monitor](https://readyforquantum.com/dashboard)
💬 **How to test**:
1. Click the **chat icon** (bottom right on any page)
2. Choose an **AI assistant type**:
- `TurboLLM` (GPT-4-mini)
- `FreeLLM` (Open-source)
- `TestLLM` (Experimental CPU-only)
### **What I’m Testing**
I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:
- **Function calling** against live network services
- **How small can a model go** while still handling:
- Automated **Nmap scans**
- **Quantum-readiness checks**
- **Metasploit integration**
🟡 **TestLLM** – Current experimental model (llama.cpp on 6 CPU threads):
- ✅ **Zero-configuration setup**
- ⏳ 30s load time (slow inference but **no API costs**)
- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!
### **Other Assistants**
🟢 **TurboLLM** – Uses **gpt-4-mini** for:
- **Real-time network diagnostics**
- **Automated penetration testing** (Nmap/Metasploit)
- 🔑 Get more tokens by [downloading our Free Network Monitor Agent](https://readyforquantum.com/download)
🔵 **HugLLM** – Open-source models (≈8B params):
- **2x more tokens** than TurboLLM
- **AI-powered log analysis**
- 🌐 Runs on Hugging Face Inference API
### 💡 **Example AI Commands to Test**:
1. `"Give me info on my websites SSL certificate"`
2. `"Check if my server is using quantum safe encyption for communication"`
3. `"Run a quick Nmap vulnerability test"`
|
reaperdoesntknow/Symiotic-14B | reaperdoesntknow | "2025-05-06T22:49:36Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen3",
"text-generation",
"symbiotic",
"symbioticai",
"llm",
"Symbols",
"conversational",
"en",
"dataset:0xZee/dataset-CoT-Advanced-Calculus-268",
"base_model:Qwen/Qwen3-14B",
"base_model:finetune:Qwen/Qwen3-14B",
"license:afl-3.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-06T21:50:24Z" | ---
license: afl-3.0
datasets:
- 0xZee/dataset-CoT-Advanced-Calculus-268
language:
- en
base_model:
- Qwen/Qwen3-14B
pipeline_tag: text-generation
library_name: transformers
tags:
- qwen3
- symbiotic
- symbioticai
- llm
- Symbols
---
# SymbioticLM-14B
**Model Type**: Hybrid Symbolic–Transformer with Persistent Memory
**Base Model**: Qwen-14B
**Framework**: PyTorch + HuggingFace Transformers
**Purpose**: Full-scale cognitive reasoning model with self-organizing memory and generative symbolic evolution
---
## Overview
SymbioticLM-14B is a state-of-the-art 17.8 billion parameter symbolic–transformer hybrid model that tightly couples high-capacity neural representation with structured symbolic cognition. Designed to match or exceed performance of top-tier LLMs in symbolic domains, it supports persistent memory, entropic recall, multi-stage symbolic routing, and self-organizing knowledge structures.
This model is ideal for advanced reasoning agents, research assistants, and symbolic math/code generation systems.
---
## Architecture Highlights
- **Backbone**: Qwen-14B transformer with rotary embeddings + FlashAttention
- **Symbolic Dim**: 8192
- **Symbolic Modules**:
- ThoughtDynamicsLNN (multi-head LSTM attention)
- LiquidThoughtProcessor
- CrystallineProcessor (DNAConv GNN)
- HelicalDNAProcessor (linear helical encoding)
- **Memory**: 4096 symbolic states in FP32, retrieved using entropy + contextual similarity
- **Dream Mode**: Background symbolic simulation for open-ended cognition
- **Router**: Intent classifier + entropy gating for processor path selection
---
## Files Included
| File | Description |
|--------------------------|----------------------------------------------------------|
| `model.bin` | Transformer weights (LFS) |
| `model.safetensors` | Memory-safe weights, optimized for loading |
| `memory.pt` | 4096-symbolic vector bank |
| `config.json` | Model and architectural metadata |
| `generation_config.json` | Top-p, temperature, decoding settings |
| `tokenizer.json` | Full tokenizer with symbolic tag support |
| `added_tokens.json` | Tags like `<D_LIM>`, `<PROOF>`, `<BY_MEASURE>`, etc. |
| `special_tokens_map.json`| Special token mapping for tokenizer |
---
## Intended Uses
- Multi-step conversational agents with true memory
- Long-form symbolic theorem generation and proof planning
- Scientific dialogue, symbolic simulations, math/code synthesis
- Reasoning in fuzzy, discontinuous, or non-smooth problem domains
---
## Limitations
- Memory requires curation and seeding for maximum utility
- Symbolic cognition is not instruction-tuned for general QA
- FlashAttention and symbolic modules increase VRAM usage during generation
---
## Citations
Please cite "SymbioticLM" when using symbolic memory components in research or applications. |
easygoing0114/Llama-3-ELYZA-JP-8B-fused | easygoing0114 | "2025-05-06T22:49:32Z" | 0 | 0 | null | [
"license:llama3.1",
"region:us"
] | null | "2025-05-04T20:44:41Z" | ---
license: llama3.1
---
Llama-3-ELYZA-JP-8B を、さまざまな AIタスク で利用しやすいように結合したモデルです。
- [オリジナルの Llama-3-ELYZA-JP-8B](https://huggingface.co/elyza/Llama-3-ELYZA-JP-8B) |
bpingua/qwen2.5_7B_geospatial_adapters | bpingua | "2025-05-06T22:49:09Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"qwen2",
"trl",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T22:48:50Z" | ---
base_model: unsloth/qwen2.5-7b-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** bpingua
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2.5-7b-unsloth-bnb-4bit
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
gvo1112/task-8-deepseek-ai-DeepSeek-R1-Distill-Qwen-14B | gvo1112 | "2025-05-06T22:47:31Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"phi3",
"custom_code",
"arxiv:1910.09700",
"base_model:microsoft/Phi-3.5-mini-instruct",
"base_model:adapter:microsoft/Phi-3.5-mini-instruct",
"region:us"
] | null | "2025-05-06T22:42:45Z" | ---
base_model: microsoft/Phi-3.5-mini-instruct
library_name: peft
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
### Framework versions
- PEFT 0.13.2 |
bprasana85/my-finetuned-TamilStory-Generator | bprasana85 | "2025-05-06T22:46:15Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T17:51:15Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
anileo1/EmpathyAI_llama3.2-3b_v3_16bit | anileo1 | "2025-05-06T22:40:39Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"llama",
"feature-extraction",
"text-generation-inference",
"unsloth",
"en",
"base_model:unsloth/Llama-3.2-1B-Instruct",
"base_model:finetune:unsloth/Llama-3.2-1B-Instruct",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | "2025-05-06T22:39:49Z" | ---
base_model: unsloth/Llama-3.2-1B-Instruct
tags:
- text-generation-inference
- transformers
- unsloth
- llama
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** anileo1
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Llama-3.2-1B-Instruct
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
gvo1112/task-8-deepseek-ai-DeepSeek-R1-Distill-Qwen-7B | gvo1112 | "2025-05-06T22:39:41Z" | 918 | 0 | peft | [
"peft",
"safetensors",
"phi3",
"custom_code",
"arxiv:1910.09700",
"base_model:microsoft/Phi-3.5-mini-instruct",
"base_model:adapter:microsoft/Phi-3.5-mini-instruct",
"region:us"
] | null | "2025-04-12T14:55:01Z" | ---
base_model: microsoft/Phi-3.5-mini-instruct
library_name: peft
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
### Framework versions
- PEFT 0.13.2 |
buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby | buelfhood | "2025-05-06T22:36:12Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T22:36:08Z" | ---
tags:
- adapter-transformers
- roberta
---
# Adapter `buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby` for microsoft/graphcodebert-base
An [adapter](https://adapterhub.ml) for the `microsoft/graphcodebert-base` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("microsoft/graphcodebert-base")
adapter_name = model.load_adapter("buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
JeloH/qwen-textgen-modelV_beni | JeloH | "2025-05-06T22:35:50Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-06T18:16:54Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
Camilla-Araujo-Go/wATCH.Camilla.Araujo.viral.video.original.link | Camilla-Araujo-Go | "2025-05-06T22:30:25Z" | 0 | 0 | null | [
"region:us"
] | null | "2025-05-06T22:27:18Z" | [🌐 CLICK HERE 🟢==►► WATCH NOW](https://videohere.top/)
[🔴 CLICK HERE 🌐==►► Download Now)](https://videohere.top/)
[<img alt="fsd" src="https://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/) |
mradermacher/MedQwen3-4B-finetuned-GGUF | mradermacher | "2025-05-06T22:28:42Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"text-generation-inference",
"unsloth",
"qwen3",
"trl",
"sft",
"en",
"base_model:ntkhoi/MedQwen3-4B-finetuned",
"base_model:quantized:ntkhoi/MedQwen3-4B-finetuned",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-06T21:11:45Z" | ---
base_model: ntkhoi/MedQwen3-4B-finetuned
language:
- en
library_name: transformers
license: apache-2.0
quantized_by: mradermacher
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
- trl
- sft
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/ntkhoi/MedQwen3-4B-finetuned
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q2_K.gguf) | Q2_K | 1.8 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q3_K_S.gguf) | Q3_K_S | 2.0 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q3_K_M.gguf) | Q3_K_M | 2.2 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q3_K_L.gguf) | Q3_K_L | 2.3 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.IQ4_XS.gguf) | IQ4_XS | 2.4 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q4_K_S.gguf) | Q4_K_S | 2.5 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q4_K_M.gguf) | Q4_K_M | 2.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q5_K_S.gguf) | Q5_K_S | 2.9 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q5_K_M.gguf) | Q5_K_M | 3.0 | |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q6_K.gguf) | Q6_K | 3.4 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.Q8_0.gguf) | Q8_0 | 4.4 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/MedQwen3-4B-finetuned-GGUF/resolve/main/MedQwen3-4B-finetuned.f16.gguf) | f16 | 8.2 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF | mradermacher | "2025-05-06T22:27:58Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"en",
"base_model:Neelectric/OLMo-2-1124-7B-Instruct_GRPOv02.04",
"base_model:quantized:Neelectric/OLMo-2-1124-7B-Instruct_GRPOv02.04",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-05-06T21:16:04Z" | ---
base_model: Neelectric/OLMo-2-1124-7B-Instruct_GRPOv02.04
language:
- en
library_name: transformers
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/Neelectric/OLMo-2-1124-7B-Instruct_GRPOv02.04
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ1_S.gguf) | i1-IQ1_S | 1.9 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ1_M.gguf) | i1-IQ1_M | 2.0 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.2 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.4 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ2_S.gguf) | i1-IQ2_S | 2.6 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q2_K_S.gguf) | i1-Q2_K_S | 2.7 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ2_M.gguf) | i1-IQ2_M | 2.8 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q2_K.gguf) | i1-Q2_K | 3.0 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 3.0 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ3_XS.gguf) | i1-IQ3_XS | 3.3 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ3_S.gguf) | i1-IQ3_S | 3.4 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.4 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ3_M.gguf) | i1-IQ3_M | 3.6 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q3_K_M.gguf) | i1-Q3_K_M | 3.8 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q3_K_L.gguf) | i1-Q3_K_L | 4.1 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ4_XS.gguf) | i1-IQ4_XS | 4.1 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-IQ4_NL.gguf) | i1-IQ4_NL | 4.3 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q4_0.gguf) | i1-Q4_0 | 4.3 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.3 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q4_K_M.gguf) | i1-Q4_K_M | 4.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q4_1.gguf) | i1-Q4_1 | 4.7 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q5_K_S.gguf) | i1-Q5_K_S | 5.2 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q5_K_M.gguf) | i1-Q5_K_M | 5.3 | |
| [GGUF](https://huggingface.co/mradermacher/OLMo-2-1124-7B-Instruct_GRPOv02.04-i1-GGUF/resolve/main/OLMo-2-1124-7B-Instruct_GRPOv02.04.i1-Q6_K.gguf) | i1-Q6_K | 6.1 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
|
Dorian2B/Vera-v0.2 | Dorian2B | "2025-05-06T22:27:37Z" | 0 | 0 | null | [
"safetensors",
"llama",
"general",
"llm",
"8B",
"text-generation",
"conversational",
"fr",
"en",
"es",
"it",
"pl",
"de",
"license:apache-2.0",
"region:us"
] | text-generation | "2025-05-06T21:25:54Z" | ---
license: apache-2.0
language:
- fr
- en
- es
- it
- pl
- de
pipeline_tag: text-generation
tags:
- general
- llm
- 8B
---
<div style="background: linear-gradient(135deg, #2d0a3e, #701f6e, #b83150); border-radius: 16px; padding: 30px; margin: 20px auto; box-shadow: 0 10px 30px rgba(0,0,0,0.3); text-align: center; max-width: 500px;">
<!-- Logo content -->
<div style="display: flex; align-items: center; justify-content: center; margin-bottom: 15px;">
<!-- Logo icon -->
<div style="width: 60px; height: 60px; background: linear-gradient(135deg, #ff8a00, #e52e71); border-radius: 12px; display: flex; align-items: center; justify-content: center; margin-right: 20px; box-shadow: 0 5px 15px rgba(229, 46, 113, 0.3);">
<span style="font-family: Arial, sans-serif; font-weight: 900; font-size: 32px; color: white; text-shadow: 0 1px 3px rgba(0,0,0,0.3);">V</span>
</div>
<!-- Logo text -->
<div style="text-align: left; position: relative;">
<div style="font-family: Arial, sans-serif; font-size: 48px; font-weight: 900; color: #ff8a00; background: linear-gradient(to right, #ff8a00, #ff5895, #cd5ff8); -webkit-background-clip: text; -webkit-text-fill-color: transparent; margin: 0; line-height: 1; letter-spacing: 1px;">VERA</div>
<div style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 500; color: #fff; background: linear-gradient(135deg, #ff8a00, #e52e71); padding: 2px 8px; border-radius: 12px; position: absolute; top: 5px; right: -40px; box-shadow: 0 3px 8px rgba(229, 46, 113, 0.3);">v0.2</div>
<div style="font-family: Arial, sans-serif; font-size: 18px; font-weight: 300; color: rgba(255,255,255,0.8); letter-spacing: 3px; text-transform: uppercase; margin: 0;">INTELLIGENCE</div>
</div>
</div>
<!-- Divider -->
<div style="width: 100%; height: 1px; background: linear-gradient(to right, transparent, rgba(255,255,255,0.5), transparent); margin: 15px auto;"></div>
<!-- Subtitle -->
<div style="font-family: Arial, sans-serif; font-size: 16px; color: rgba(255,255,255,0.7); font-style: italic; margin-top: 12px;">Solution d'intelligence artificielle nouvelle génération</div>
</div>
# Vera v0.2
**Créé le :** 7 mai 2025
**Auteur :** Dorian Dominici
**Paramètres :** 8 milliards
**Contexte max. :** 128 000 tokens
---
## 🌟 Description
Vera est un modèle de langage polyvalent (LLM) multilingue, conçu pour offrir un **échange naturel** principalement en **français** et en **anglais**, avec un support secondaire pour l'**espagnol**, l'**italien**, l'**allemand** et le **polonais**. Grâce à ses 8 milliards de paramètres et à une fenêtre contextuelle considérablement étendue à 128 k tokens, Vera excelle dans :
- 💬 **Conversation fluide et naturelle**
- 🔄 **Traduction précise et contextuelle**
- 📝 **Génération et correction de code avancées**
- 🤖 **Agents IA** pour tâches complexes
- 📊 **Analyse de documents volumineux**
---
## 🚀 Points forts
- **Multilingue** : Excellence en français et anglais, avec support solide pour l'espagnol, l'italien, l'allemand et le polonais.
- **Contexte étendu** : Fenêtre de 128k tokens idéale pour l'analyse de longs documents et scénarios d'agents IA complexes.
- **Connaissance générale élevée** : Base de connaissances étendue couvrant un large éventail de domaines académiques, culturels et pratiques.
- **Polyvalence améliorée** : Performances supérieures en chat, traduction, résumé, codage et raisonnement.
- **Compétences techniques** : Très bonnes aptitudes en programmation, analyse de données et rédaction technique.
- **Accès open-source** : Facilement déployable et intégrable via la plateforme Hugging Face.
---
## 🧱 Points d'amélioration
- **Spécialisation** : Bien que polyvalent, peut être moins performant que des modèles spécialisés pour certaines tâches très spécifiques.
- **Taille modérée** : Avec 8 milliards de paramètres, reste plus compact que les modèles géants (tout en offrant un excellent rapport performances/ressources).
---
## 🛠️ Cas d'usage
| Domaine | Exemples |
|------------------------|-------------------------------------------------|
| Chatbot & Assistance | Support client multilingue, systèmes conversationnels avancés |
| Traduction | Textes techniques, documentation spécialisée, littérature |
| Développement logiciel | Génération de code, débogage, documentation automatisée |
| Rédaction & Analyse | Articles, rapports, synthèses de documents volumineux |
| Automatisation IA | Agents conversationnels complexes, systèmes de RAG |
| Éducation | Tutoriels personnalisés, assistance à l'apprentissage |
---
## 📦 Détails techniques
- **Architecture** : Transformer optimisé
- **Taille du modèle** : 8 milliards de paramètres
- **Context window** : 128 000 tokens
- **Langues principales** : Français, Anglais
- **Langues secondaires** : Espagnol, Italien, Allemand, Polonais
- **Licence** : Apache-2.0
--- |
augustocsc/Se124M500KInfDelimiter | augustocsc | "2025-05-06T22:26:40Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"generated_from_trainer",
"base_model:openai-community/gpt2",
"base_model:adapter:openai-community/gpt2",
"license:mit",
"region:us"
] | null | "2025-05-06T21:40:08Z" | ---
library_name: peft
license: mit
base_model: gpt2
tags:
- generated_from_trainer
model-index:
- name: Se124M500KInfDelimiter
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# Se124M500KInfDelimiter
This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5132
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 3
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:-----:|:---------------:|
| 0.1376 | 1.0 | 7890 | 0.5302 |
| 0.1321 | 2.0 | 15780 | 0.5167 |
| 0.131 | 3.0 | 23670 | 0.5132 |
### Framework versions
- PEFT 0.15.1
- Transformers 4.51.3
- Pytorch 2.6.0+cu118
- Datasets 3.5.0
- Tokenizers 0.21.1 |
seohuibae/fc-sft-full-llama3-alf-dpo-iter-1 | seohuibae | "2025-05-06T22:26:34Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-05-06T22:17:18Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
buelfhood/conplag1_codebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby | buelfhood | "2025-05-06T22:26:33Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T22:26:30Z" | ---
tags:
- roberta
- adapter-transformers
---
# Adapter `buelfhood/conplag1_codebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby` for microsoft/codebert-base
An [adapter](https://adapterhub.ml) for the `microsoft/codebert-base` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("microsoft/codebert-base")
adapter_name = model.load_adapter("buelfhood/conplag1_codebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_houlsby", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
KSandke/legal-simplifier | KSandke | "2025-05-06T22:25:43Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"pegasus",
"text2text-generation",
"arxiv:1910.09700",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | "2025-05-06T22:05:05Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
kumasi-video-go/wATCH.kumasi.video.viral.video.original | kumasi-video-go | "2025-05-06T22:25:42Z" | 0 | 0 | null | [
"region:us"
] | null | "2025-05-06T22:23:45Z" | [🌐 CLICK HERE 🟢==►► WATCH NOW](https://videohere.top/)
[🔴 CLICK HERE 🌐==►► Download Now)](https://videohere.top/)
[<img alt="fsd" src="https://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/) |
epitta/llama3.2vision_fine-tuning_1epoch_with_explanations | epitta | "2025-05-06T22:23:43Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"mllama",
"trl",
"en",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T22:23:28Z" | ---
base_model: unsloth/llama-3.2-11b-vision-instruct-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- mllama
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** epitta
- **License:** apache-2.0
- **Finetuned from model :** unsloth/llama-3.2-11b-vision-instruct-unsloth-bnb-4bit
This mllama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
breadlicker45/breadchat-save2000 | breadlicker45 | "2025-05-06T22:20:36Z" | 4 | 0 | null | [
"safetensors",
"granite",
"dataset:breadlicker45/bread-chat-sft",
"base_model:ibm-granite/granite-3.3-2b-base",
"base_model:finetune:ibm-granite/granite-3.3-2b-base",
"region:us"
] | null | "2025-05-03T20:18:18Z" | ---
base_model:
- ibm-granite/granite-3.3-2b-base
datasets:
- breadlicker45/bread-chat-sft
--- |
averntech/Avern-ARC-1X | averntech | "2025-05-06T22:20:18Z" | 0 | 0 | null | [
"pytorch",
"mistral",
"dataset:philschmid/guanaco-sharegpt-style",
"base_model:mistralai/Mistral-7B-Instruct-v0.3",
"base_model:finetune:mistralai/Mistral-7B-Instruct-v0.3",
"license:mit",
"region:us"
] | null | "2025-05-06T15:49:12Z" | ---
license: mit
datasets:
- philschmid/guanaco-sharegpt-style
metrics:
- accuracy
- character
base_model:
- mistralai/Mistral-7B-Instruct-v0.3
--- |
mradermacher/QRWKV6-7B-Instruct-GGUF | mradermacher | "2025-05-06T22:19:39Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"en",
"base_model:recursal/QRWKV6-7B-Instruct",
"base_model:quantized:recursal/QRWKV6-7B-Instruct",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-06T21:27:23Z" | ---
base_model: recursal/QRWKV6-7B-Instruct
language:
- en
library_name: transformers
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/recursal/QRWKV6-7B-Instruct
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q2_K.gguf) | Q2_K | 3.7 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q3_K_S.gguf) | Q3_K_S | 4.2 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q3_K_M.gguf) | Q3_K_M | 4.5 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q3_K_L.gguf) | Q3_K_L | 4.7 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.IQ4_XS.gguf) | IQ4_XS | 5.0 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q4_K_S.gguf) | Q4_K_S | 5.2 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q4_K_M.gguf) | Q4_K_M | 5.4 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q5_K_S.gguf) | Q5_K_S | 6.1 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q5_K_M.gguf) | Q5_K_M | 6.3 | |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q6_K.gguf) | Q6_K | 7.1 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.Q8_0.gguf) | Q8_0 | 9.0 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/QRWKV6-7B-Instruct-GGUF/resolve/main/QRWKV6-7B-Instruct.f16.gguf) | f16 | 16.5 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
yanmingzhu/videomae-base-finetuned-ucf101-subset | yanmingzhu | "2025-05-06T22:19:37Z" | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"videomae",
"video-classification",
"generated_from_trainer",
"base_model:MCG-NJU/videomae-base",
"base_model:finetune:MCG-NJU/videomae-base",
"license:cc-by-nc-4.0",
"endpoints_compatible",
"region:us"
] | video-classification | "2025-05-06T21:37:01Z" | ---
library_name: transformers
license: cc-by-nc-4.0
base_model: MCG-NJU/videomae-base
tags:
- generated_from_trainer
metrics:
- accuracy
model-index:
- name: videomae-base-finetuned-ucf101-subset
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# videomae-base-finetuned-ucf101-subset
This model is a fine-tuned version of [MCG-NJU/videomae-base](https://huggingface.co/MCG-NJU/videomae-base) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4468
- Accuracy: 0.8286
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- training_steps: 148
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:------:|:----:|:---------------:|:--------:|
| 2.147 | 0.2568 | 38 | 1.8615 | 0.4857 |
| 0.8927 | 1.2568 | 76 | 0.9193 | 0.6286 |
| 0.4778 | 2.2568 | 114 | 0.5483 | 0.8571 |
| 0.2702 | 3.2297 | 148 | 0.4468 | 0.8286 |
### Framework versions
- Transformers 4.51.3
- Pytorch 2.7.0+cu126
- Datasets 3.5.1
- Tokenizers 0.21.1
|
jackal79/tle-orbit-explainer | jackal79 | "2025-05-06T22:16:51Z" | 9 | 0 | peft | [
"peft",
"safetensors",
"LoRA",
"TLE",
"space-domain-awareness",
"trajectory-prediction",
"orbital-mechanics",
"text-generation",
"conversational",
"base_model:Qwen/Qwen1.5-7B",
"base_model:adapter:Qwen/Qwen1.5-7B",
"license:other",
"region:us"
] | text-generation | "2025-05-05T02:42:38Z" | ---
pipeline_tag: text-generation
base_model: Qwen/Qwen1.5-7B
library_name: peft
tags:
- LoRA
- TLE
- space-domain-awareness
- trajectory-prediction
- orbital-mechanics
license: other
---
# tle-orbit-explainer
A LoRA adapter for **Qwen-1.5-7B** that translates raw Two-Line Elements (TLEs) into natural-language orbit explanations, decay risk scores, and anomaly flags for general space awareness workflows.
---
## Model Details
### Model Description
| | |
| ------------------ | ----------------------------------------------------------- |
| **Developed by** | Jack Al-Kahwati / Stardrive |
| **Funded by** | ⬜️ (Self-funded) |
| **Shared by** | jackal79 (Hugging Face) |
| **Model type** | LoRA adapter (`peft==0.10.0`) |
| **Languages** | English |
| **License** | TLE-Orbit-NonCommercial v1.0 ([custom terms](./LICENSE.txt)) |
| **Finetuned from** | [`Qwen/Qwen1.5-7B`](https://huggingface.co/Qwen/Qwen1.5-7B) |
### Model Sources
| | |
| ---------------- | ---------------------------------------------------------------------------------------------------------- |
| **Repository** | [https://huggingface.co/jackal79/tle-orbit-explainer](https://huggingface.co/jackal79/tle-orbit-explainer) |
| **Paper / Blog** | https://medium.com/@jack_16944/enhancing-space-awareness-with-fine-tuned-transformer-models-introducing-tle-orbit-explainer-67ae40653ed5 |
---
## Uses
### Direct Use
* Quick summarization of satellite orbital states for analysts
* Plain-language TLE explanations for educational purposes
* Offline dataset labeling (orbital classifications)
### Downstream Use
* Combine with SGP4 for enhanced position forecasting
* Integration into satellite autonomy stacks (cubesats, small-scale hardware)
* Pre-prompted agent support in secure orbital management workflows
### Out-of-Scope Use
* High-precision orbit propagation without additional physics modeling
* Applications related to targeting, weapons systems, or lethal autonomous decisions
* Jurisdictions prohibiting ML or data export (verify with ITAR/EAR guidelines)
---
## Bias, Risks, & Limitations
| Category | Note |
| ------------------- | ------------------------------------------------------------------------------------------------------------- |
| **Data bias** | Trained primarily on decayed objects (`DECAY = 1`), possibly underestimating longevity for active satellites. |
| **Temporal limits** | Operates on snapshot data; does not handle continuous high-frequency time-series. |
| **Language** | Supports explanations in English only. |
| **Accuracy** | Potential inaccuracies in decay date predictions; verify independently. |
### Recommendations
Incorporate independent physics-based validation before operational use and maintain a human-in-the-loop for any critical or high-risk decisions.
---
## How to Get Started
```python
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from peft import PeftModel
base = "Qwen/Qwen1.5-7B"
lora = "jackal79/tle-orbit-explainer"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, lora) # merges LoRA
pipe = pipeline("text-generation", model=model, tokenizer=tok, device=0)
prompt = """### Prompt:
1 25544U 98067A 24079.07757601 .00016717 00000+0 10270-3 0 9994
2 25544 51.6400 337.6640 0007776 35.5310 330.5120 15.50377579499263
### Reasoning:
"""
print(pipe(prompt, max_new_tokens=120)[0]["generated_text"])
```
---
## License
This model is released under the **TLE-Orbit-NonCommercial License v1.0**.
- ✅ Free for non-commercial use, research, and internal evaluation
- 🚫 Commercial, operational, or for-profit use requires a separate license
To request a commercial license, contact: [email protected]
|
KSandke/legal-classifier | KSandke | "2025-05-06T22:11:32Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"bert",
"text-classification",
"arxiv:1910.09700",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | "2025-05-06T21:36:03Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
Egika/sooz | Egika | "2025-05-06T22:09:18Z" | 0 | 0 | null | [
"license:apache-2.0",
"region:us"
] | null | "2025-04-13T10:14:56Z" | ---
license: apache-2.0
---
|
mradermacher/alfworld-1.5b-lora-GGUF | mradermacher | "2025-05-06T22:07:57Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"en",
"base_model:Xiaofeng77/alfworld-1.5b-lora",
"base_model:quantized:Xiaofeng77/alfworld-1.5b-lora",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-06T21:57:53Z" | ---
base_model: Xiaofeng77/alfworld-1.5b-lora
language:
- en
library_name: transformers
quantized_by: mradermacher
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/Xiaofeng77/alfworld-1.5b-lora
<!-- provided-files -->
weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion.
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q2_K.gguf) | Q2_K | 0.8 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q3_K_S.gguf) | Q3_K_S | 0.9 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q3_K_M.gguf) | Q3_K_M | 0.9 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q3_K_L.gguf) | Q3_K_L | 1.0 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.IQ4_XS.gguf) | IQ4_XS | 1.0 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q4_K_S.gguf) | Q4_K_S | 1.0 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q4_K_M.gguf) | Q4_K_M | 1.1 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q5_K_S.gguf) | Q5_K_S | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q5_K_M.gguf) | Q5_K_M | 1.2 | |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q6_K.gguf) | Q6_K | 1.4 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.Q8_0.gguf) | Q8_0 | 1.7 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/alfworld-1.5b-lora-GGUF/resolve/main/alfworld-1.5b-lora.f16.gguf) | f16 | 3.2 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
mxmcc/xLAM-2-32b-fc-r-4bit | mxmcc | "2025-05-06T22:03:05Z" | 0 | 0 | mlx | [
"mlx",
"safetensors",
"qwen2",
"function-calling",
"LLM Agent",
"tool-use",
"llama",
"qwen",
"pytorch",
"LLaMA-factory",
"text-generation",
"conversational",
"en",
"dataset:Salesforce/xlam-function-calling-60k",
"base_model:Salesforce/xLAM-2-32b-fc-r",
"base_model:quantized:Salesforce/xLAM-2-32b-fc-r",
"license:cc-by-nc-4.0",
"4-bit",
"region:us"
] | text-generation | "2025-05-06T21:58:46Z" | ---
license: cc-by-nc-4.0
datasets:
- Salesforce/xlam-function-calling-60k
language:
- en
pipeline_tag: text-generation
tags:
- function-calling
- LLM Agent
- tool-use
- llama
- qwen
- pytorch
- LLaMA-factory
- mlx
library_name: mlx
base_model: Salesforce/xLAM-2-32b-fc-r
---
|
pt4c/marian-finetuned-kde4-en-to-ig | pt4c | "2025-05-06T22:02:56Z" | 0 | 0 | transformers | [
"transformers",
"tf",
"marian",
"text2text-generation",
"generated_from_keras_callback",
"base_model:Helsinki-NLP/opus-mt-en-ig",
"base_model:finetune:Helsinki-NLP/opus-mt-en-ig",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | "2025-05-06T22:01:36Z" | ---
library_name: transformers
license: apache-2.0
base_model: Helsinki-NLP/opus-mt-en-ig
tags:
- generated_from_keras_callback
model-index:
- name: pt4c/marian-finetuned-kde4-en-to-ig
results: []
---
<!-- This model card has been generated automatically according to the information Keras had access to. You should
probably proofread and complete it, then remove this comment. -->
# pt4c/marian-finetuned-kde4-en-to-ig
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-en-ig](https://huggingface.co/Helsinki-NLP/opus-mt-en-ig) on an unknown dataset.
It achieves the following results on the evaluation set:
- Train Loss: 3.9965
- Validation Loss: 4.5130
- Epoch: 2
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'module': 'keras.optimizers.schedules', 'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-05, 'decay_steps': 93, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, 'registered_name': None}, 'decay': 0.0, 'beta_1': np.float32(0.9), 'beta_2': np.float32(0.999), 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
- training_precision: mixed_float16
### Training results
| Train Loss | Validation Loss | Epoch |
|:----------:|:---------------:|:-----:|
| 4.8171 | 4.6366 | 0 |
| 4.2146 | 4.5421 | 1 |
| 3.9965 | 4.5130 | 2 |
### Framework versions
- Transformers 4.51.3
- TensorFlow 2.18.0
- Datasets 3.5.1
- Tokenizers 0.21.1
|
buelfhood/conplag1_codet5_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora | buelfhood | "2025-05-06T22:02:38Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"t5",
"region:us"
] | null | "2025-05-06T22:02:35Z" | ---
tags:
- t5
- adapter-transformers
---
# Adapter `buelfhood/conplag1_codet5_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora` for Salesforce/codet5-small
An [adapter](https://adapterhub.ml) for the `Salesforce/codet5-small` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("Salesforce/codet5-small")
adapter_name = model.load_adapter("buelfhood/conplag1_codet5_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
infogeo/fd8aeb85-7ad1-4d67-ad16-c2f969bfa8cb | infogeo | "2025-05-06T21:56:35Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"gpt_neox",
"axolotl",
"generated_from_trainer",
"base_model:EleutherAI/pythia-1b",
"base_model:adapter:EleutherAI/pythia-1b",
"license:apache-2.0",
"4-bit",
"bitsandbytes",
"region:us"
] | null | "2025-05-06T21:52:39Z" | ---
library_name: peft
license: apache-2.0
base_model: EleutherAI/pythia-1b
tags:
- axolotl
- generated_from_trainer
model-index:
- name: fd8aeb85-7ad1-4d67-ad16-c2f969bfa8cb
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.4.1`
```yaml
absolute_data_files: false
adapter: lora
base_model: EleutherAI/pythia-1b
bf16: true
chat_template: llama3
dataset_prepared_path: /workspace/axolotl
datasets:
- data_files:
- 1703dbe60f29fbac_train_data.json
ds_type: json
format: custom
path: /workspace/input_data/1703dbe60f29fbac_train_data.json
type:
field_input: input
field_instruction: instruction
field_output: output
format: '{instruction} {input}'
no_input_format: '{instruction}'
system_format: '{system}'
system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 1
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_clipping: 0.55
group_by_length: false
hub_model_id: infogeo/fd8aeb85-7ad1-4d67-ad16-c2f969bfa8cb
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 1.0e-06
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
max_steps: 400
micro_batch_size: 8
mixed_precision: bf16
mlflow_experiment_name: /tmp/1703dbe60f29fbac_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 1
sequence_len: 1024
special_tokens:
pad_token: <|endoftext|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 289a62f8-96f0-4ebe-abc7-f59e0a56c6b5
wandb_project: s56-28
wandb_run: your_name
wandb_runid: 289a62f8-96f0-4ebe-abc7-f59e0a56c6b5
warmup_steps: 20
weight_decay: 0.01
xformers_attention: true
```
</details><br>
# fd8aeb85-7ad1-4d67-ad16-c2f969bfa8cb
This model is a fine-tuned version of [EleutherAI/pythia-1b](https://huggingface.co/EleutherAI/pythia-1b) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 3.3101
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-06
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 20
- training_steps: 400
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 3.1367 | 0.0303 | 400 | 3.3101 |
### Framework versions
- PEFT 0.13.2
- Transformers 4.46.0
- Pytorch 2.5.0+cu124
- Datasets 3.0.1
- Tokenizers 0.20.1 |
dulimov/Qwen3-0.6B-rk3588-1.2.1-unsloth | dulimov | "2025-05-06T21:56:31Z" | 1 | 0 | null | [
"qwen3",
"unsloth",
"base_model:Qwen/Qwen3-0.6B",
"base_model:finetune:Qwen/Qwen3-0.6B",
"region:us"
] | null | "2025-05-05T12:32:15Z" | ---
base_model:
- Qwen/Qwen3-0.6B
tags:
- unsloth
---
# Qwen3-0.6B-RK3588-1.2.1
This version of Qwen3-0.6B has been converted to run on the RK3588 NPU using ['w8a8', 'w8a8_g128', 'w8a8_g256', 'w8a8_g512'] quantization.
This model has been optimized with the following LoRA:
Compatible with RKLLM version: 1.2.1
# Original Model Card for base model, Qwen3-0.6B, below:
# Qwen3-0.6B
## Qwen3 Highlights
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features:
- **Uniquely support of seamless switching between thinking mode** (for complex logical reasoning, math, and coding) and **non-thinking mode** (for efficient, general-purpose dialogue) **within single model**, ensuring optimal performance across various scenarios.
- **Significantly enhancement in its reasoning capabilities**, surpassing previous QwQ (in thinking mode) and Qwen2.5 instruct models (in non-thinking mode) on mathematics, code generation, and commonsense logical reasoning.
- **Superior human preference alignment**, excelling in creative writing, role-playing, multi-turn dialogues, and instruction following, to deliver a more natural, engaging, and immersive conversational experience.
- **Expertise in agent capabilities**, enabling precise integration with external tools in both thinking and unthinking modes and achieving leading performance among open-source models in complex agent-based tasks.
- **Support of 100+ languages and dialects** with strong capabilities for **multilingual instruction following** and **translation**.
## Model Overview
**Qwen3-0.6B** has the following features:
- Type: Causal Language Models
- Training Stage: Pretraining & Post-training
- Number of Parameters: 0.6B
- Number of Paramaters (Non-Embedding): 0.44B
- Number of Layers: 28
- Number of Attention Heads (GQA): 16 for Q and 8 for KV
- Context Length: 32,768
For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our [blog](https://qwenlm.github.io/blog/qwen3/), [GitHub](https://github.com/QwenLM/Qwen3), and [Documentation](https://qwen.readthedocs.io/en/latest/).
## Quickstart
The code of Qwen3 has been in the latest Hugging Face `transformers` and we advise you to use the latest version of `transformers`.
With `transformers<4.51.0`, you will encounter the following error:
```
KeyError: 'qwen3'
```
The following contains a code snippet illustrating how to use the model generate content based on given inputs.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "Qwen/Qwen3-0.6B"
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
# conduct text completion
generated_ids = model.generate(
**model_inputs,
max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
# parsing thinking content
try:
# rindex finding 151668 (</think>)
index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
index = 0
thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
print("thinking content:", thinking_content)
print("content:", content)
```
For deployment, you can use `vllm>=0.8.5` or `sglang>=0.4.5.post2` to create an OpenAI-compatible API endpoint:
- vLLM:
```shell
vllm serve Qwen/Qwen3-0.6B --enable-reasoning --reasoning-parser deepseek_r1
```
- SGLang:
```shell
python -m sglang.launch_server --model-path Qwen/Qwen3-0.6B --reasoning-parser deepseek-r1
```
## Switching Between Thinking and Non-Thinking Mode
> [!TIP]
> The `enable_thinking` switch is also available in APIs created by vLLM and SGLang.
> Please refer to [our documentation](https://qwen.readthedocs.io/) for more details.
### `enable_thinking=True`
By default, Qwen3 has thinking capabilities enabled, similar to QwQ-32B. This means the model will use its reasoning abilities to enhance the quality of generated responses. For example, when explicitly setting `enable_thinking=True` or leaving it as the default value in `tokenizer.apply_chat_template`, the model will engage its thinking mode.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # True is the default value for enable_thinking
)
```
In this mode, the model will generate think content wrapped in a `<think>...</think>` block, followed by the final response.
> [!NOTE]
> For thinking mode, use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0` (the default setting in `generation_config.json`). **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### `enable_thinking=False`
We provide a hard switch to strictly disable the model's thinking behavior, aligning its functionality with the previous Qwen2.5-Instruct models. This mode is particularly useful in scenarios where disabling thinking is essential for enhancing efficiency.
```python
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False # Setting enable_thinking=False disables thinking mode
)
```
In this mode, the model will not generate any think content and will not include a `<think>...</think>` block.
> [!NOTE]
> For non-thinking mode, we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`. For more detailed guidance, please refer to the [Best Practices](#best-practices) section.
### Advanced Usage: Switching Between Thinking and Non-Thinking Modes via User Input
We provide a soft switch mechanism that allows users to dynamically control the model's behavior when `enable_thinking=True`. Specifically, you can add `/think` and `/no_think` to user prompts or system messages to switch the model's thinking mode from turn to turn. The model will follow the most recent instruction in multi-turn conversations.
Here is an example of a multi-turn conversation:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
class QwenChatbot:
def __init__(self, model_name="Qwen/Qwen3-0.6B"):
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
self.model = AutoModelForCausalLM.from_pretrained(model_name)
self.history = []
def generate_response(self, user_input):
messages = self.history + [{"role": "user", "content": user_input}]
text = self.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
inputs = self.tokenizer(text, return_tensors="pt")
response_ids = self.model.generate(**inputs, max_new_tokens=32768)[0][len(inputs.input_ids[0]):].tolist()
response = self.tokenizer.decode(response_ids, skip_special_tokens=True)
# Update history
self.history.append({"role": "user", "content": user_input})
self.history.append({"role": "assistant", "content": response})
return response
# Example Usage
if __name__ == "__main__":
chatbot = QwenChatbot()
# First input (without /think or /no_think tags, thinking mode is enabled by default)
user_input_1 = "How many r's in strawberries?"
print(f"User: {user_input_1}")
response_1 = chatbot.generate_response(user_input_1)
print(f"Bot: {response_1}")
print("----------------------")
# Second input with /no_think
user_input_2 = "Then, how many r's in blueberries? /no_think"
print(f"User: {user_input_2}")
response_2 = chatbot.generate_response(user_input_2)
print(f"Bot: {response_2}")
print("----------------------")
# Third input with /think
user_input_3 = "Really? /think"
print(f"User: {user_input_3}")
response_3 = chatbot.generate_response(user_input_3)
print(f"Bot: {response_3}")
```
> **Note**
> For API compatibility, when `enable_thinking=True`, regardless of whether the user uses `/think` or `/no_think`, the model will always output a block wrapped in `<think>...</think>`. However, the content inside this block may be empty if thinking is disabled.
> When `enable_thinking=False`, the soft switches are not valid. Regardless of any `/think` or `/no_think` tags input by the user, the model will not generate think content and will not include a `<think>...</think>` block.
## Agentic Use
Qwen3 excels in tool calling capabilities. We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to make the best use of agentic ability of Qwen3. Qwen-Agent encapsulates tool-calling templates and tool-calling parsers internally, greatly reducing coding complexity.
To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
```python
from qwen_agent.agents import Assistant
# Define LLM
llm_cfg = {
'model': 'Qwen3-0.6B',
# Use the endpoint provided by Alibaba Model Studio:
# 'model_type': 'qwen_dashscope',
# 'api_key': os.getenv('DASHSCOPE_API_KEY'),
# Use a custom endpoint compatible with OpenAI API:
'model_server': 'http://localhost:8000/v1', # api_base
'api_key': 'EMPTY',
# Other parameters:
# 'generate_cfg': {
# # Add: When the response content is `<think>this is the thought</think>this is the answer;
# # Do not add: When the response has been separated by reasoning_content and content.
# 'thought_in_content': True,
# },
}
# Define Tools
tools = [
{'mcpServers': { # You can specify the MCP configuration file
'time': {
'command': 'uvx',
'args': ['mcp-server-time', '--local-timezone=Asia/Shanghai']
},
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
},
'code_interpreter', # Built-in tools
]
# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)
# Streaming generation
messages = [{'role': 'user', 'content': 'https://qwenlm.github.io/blog/ Introduce the latest developments of Qwen'}]
for responses in bot.run(messages=messages):
pass
print(responses)
```
## Best Practices
To achieve optimal performance, we recommend the following settings:
1. **Sampling Parameters**:
- For thinking mode (`enable_thinking=True`), use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0`. **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions.
- For non-thinking mode (`enable_thinking=False`), we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`.
- For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 38,912 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
- **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
- **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
4. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
### Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{qwen3,
title = {Qwen3},
url = {https://qwenlm.github.io/blog/qwen3/},
author = {Qwen Team},
month = {April},
year = {2025}
}
``` |
buelfhood/conplag1_codeberta_ep50_bs16_lr0_0005_l512_s42_ppy_f_beta_score_lora | buelfhood | "2025-05-06T21:56:24Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T21:56:22Z" | ---
tags:
- adapter-transformers
- roberta
---
# Adapter `buelfhood/conplag1_codeberta_ep50_bs16_lr0_0005_l512_s42_ppy_f_beta_score_lora` for huggingface/CodeBERTa-small-v1
An [adapter](https://adapterhub.ml) for the `huggingface/CodeBERTa-small-v1` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("huggingface/CodeBERTa-small-v1")
adapter_name = model.load_adapter("buelfhood/conplag1_codeberta_ep50_bs16_lr0_0005_l512_s42_ppy_f_beta_score_lora", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
azigi-ooh-azigi-go/wATCH.azigi.ooh.azigi.viral.video.original | azigi-ooh-azigi-go | "2025-05-06T21:55:49Z" | 0 | 0 | null | [
"region:us"
] | null | "2025-05-06T21:54:04Z" | [🌐 CLICK HERE 🟢==►► WATCH NOW](https://videohere.top/)
[🔴 CLICK HERE 🌐==►► Download Now)](https://videohere.top/)
[<img alt="fsd" src="https://i.postimg.cc/qvPp49Sm/ythngythg.gif">](https://videohere.top/) |
buelfhood/conplag1_codeberta_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora | buelfhood | "2025-05-06T21:51:41Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T21:51:39Z" | ---
tags:
- roberta
- adapter-transformers
---
# Adapter `buelfhood/conplag1_codeberta_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora` for huggingface/CodeBERTa-small-v1
An [adapter](https://adapterhub.ml) for the `huggingface/CodeBERTa-small-v1` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("huggingface/CodeBERTa-small-v1")
adapter_name = model.load_adapter("buelfhood/conplag1_codeberta_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
bprhee01/bert-text-classification | bprhee01 | "2025-05-06T21:50:44Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"bert",
"text-classification",
"generated_from_trainer",
"base_model:google-bert/bert-base-uncased",
"base_model:finetune:google-bert/bert-base-uncased",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | "2025-05-06T21:09:28Z" | ---
library_name: transformers
license: apache-2.0
base_model: bert-base-uncased
tags:
- generated_from_trainer
model-index:
- name: bert-text-classification
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# bert-text-classification
This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on an unknown dataset.
It achieves the following results on the evaluation set:
- eval_loss: 0.0356
- eval_accuracy: 0.992
- eval_f1: 0.9892
- eval_precision: 0.9829
- eval_recall: 0.9957
- eval_runtime: 34.7211
- eval_samples_per_second: 144.005
- eval_steps_per_second: 2.275
- epoch: 1.8648
- step: 1600
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 5
- mixed_precision_training: Native AMP
### Framework versions
- Transformers 4.51.3
- Pytorch 2.7.0+cu126
- Datasets 3.5.1
- Tokenizers 0.21.1
|
mahsharyahan/EMBEDDIA_crosloengual_bert_Sl | mahsharyahan | "2025-05-06T21:48:42Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"bert",
"text-classification",
"generated_from_trainer",
"base_model:EMBEDDIA/crosloengual-bert",
"base_model:finetune:EMBEDDIA/crosloengual-bert",
"license:cc-by-4.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | "2025-05-06T21:48:25Z" | ---
library_name: transformers
license: cc-by-4.0
base_model: EMBEDDIA/crosloengual-bert
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
- precision
- recall
model-index:
- name: EMBEDDIA_crosloengual_bert_Sl
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# EMBEDDIA_crosloengual_bert_Sl
This model is a fine-tuned version of [EMBEDDIA/crosloengual-bert](https://huggingface.co/EMBEDDIA/crosloengual-bert) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5078
- Accuracy: 0.75
- F1: 0.8182
- Precision: 1.0
- Recall: 0.6923
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 5
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | Precision | Recall |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:---------:|:------:|
| No log | 1.0 | 13 | 0.2552 | 0.9375 | 0.9600 | 1.0 | 0.9231 |
| No log | 2.0 | 26 | 0.4370 | 0.75 | 0.8182 | 1.0 | 0.6923 |
| No log | 3.0 | 39 | 0.4171 | 0.8125 | 0.8696 | 1.0 | 0.7692 |
| No log | 4.0 | 52 | 0.3939 | 0.8125 | 0.8696 | 1.0 | 0.7692 |
| No log | 5.0 | 65 | 0.5078 | 0.75 | 0.8182 | 1.0 | 0.6923 |
### Framework versions
- Transformers 4.51.1
- Pytorch 2.5.1+cu124
- Datasets 3.5.0
- Tokenizers 0.21.0
|
mradermacher/Qwen3-30B-A3B-python-coder-i1-GGUF | mradermacher | "2025-05-06T21:45:16Z" | 0 | 0 | null | [
"gguf",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-05-06T17:06:25Z" | <!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/burtenshaw/Qwen3-30B-A3B-python-coder
|
RaghuCourage9605/Custom_LLM | RaghuCourage9605 | "2025-05-06T21:44:57Z" | 3 | 0 | null | [
"license:apache-2.0",
"region:us"
] | null | "2025-04-26T11:40:25Z" | ---
license: apache-2.0
---
|
yesbreaddog/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-peaceful_graceful_llama | yesbreaddog | "2025-05-06T21:41:44Z" | 1 | 0 | transformers | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"generated_from_trainer",
"rl-swarm",
"grpo",
"gensyn",
"I am peaceful graceful llama",
"trl",
"conversational",
"arxiv:2402.03300",
"base_model:Gensyn/Qwen2.5-0.5B-Instruct",
"base_model:finetune:Gensyn/Qwen2.5-0.5B-Instruct",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | "2025-04-27T23:14:10Z" | ---
base_model: Gensyn/Qwen2.5-0.5B-Instruct
library_name: transformers
model_name: Qwen2.5-0.5B-Instruct-Gensyn-Swarm-peaceful_graceful_llama
tags:
- generated_from_trainer
- rl-swarm
- grpo
- gensyn
- I am peaceful graceful llama
- trl
licence: license
---
# Model Card for Qwen2.5-0.5B-Instruct-Gensyn-Swarm-peaceful_graceful_llama
This model is a fine-tuned version of [Gensyn/Qwen2.5-0.5B-Instruct](https://huggingface.co/Gensyn/Qwen2.5-0.5B-Instruct).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="yesbreaddog/Qwen2.5-0.5B-Instruct-Gensyn-Swarm-peaceful_graceful_llama", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
### Framework versions
- TRL: 0.15.2
- Transformers: 4.51.3
- Pytorch: 2.7.0
- Datasets: 3.5.0
- Tokenizers: 0.21.1
## Citations
Cite GRPO as:
```bibtex
@article{zhihong2024deepseekmath,
title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
year = 2024,
eprint = {arXiv:2402.03300},
}
```
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
marialvsantiago/e083b4d3-573f-41e9-8468-fbc799e27adf | marialvsantiago | "2025-05-06T21:36:48Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"mistral",
"axolotl",
"generated_from_trainer",
"base_model:teknium/OpenHermes-2.5-Mistral-7B",
"base_model:adapter:teknium/OpenHermes-2.5-Mistral-7B",
"license:apache-2.0",
"4-bit",
"bitsandbytes",
"region:us"
] | null | "2025-05-06T21:28:44Z" | ---
library_name: peft
license: apache-2.0
base_model: teknium/OpenHermes-2.5-Mistral-7B
tags:
- axolotl
- generated_from_trainer
model-index:
- name: e083b4d3-573f-41e9-8468-fbc799e27adf
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.4.1`
```yaml
adapter: lora
base_model: teknium/OpenHermes-2.5-Mistral-7B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
- 0028f9871f835ea6_train_data.json
ds_type: json
format: custom
path: /workspace/input_data/0028f9871f835ea6_train_data.json
type:
field_input: input
field_instruction: instruction
field_output: output
format: '{instruction} {input}'
no_input_format: '{instruction}'
system_format: '{system}'
system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 1
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_clipping: 0.5
group_by_length: false
hub_model_id: marialvsantiago/e083b4d3-573f-41e9-8468-fbc799e27adf
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 5.0e-06
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
max_steps: 350
micro_batch_size: 8
mixed_precision: bf16
mlflow_experiment_name: /tmp/0028f9871f835ea6_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 1
sequence_len: 1024
special_tokens:
pad_token: <|im_end|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 14285fca-fbf7-4ffd-a920-2a363c95d04d
wandb_project: s56-33
wandb_run: your_name
wandb_runid: 14285fca-fbf7-4ffd-a920-2a363c95d04d
warmup_steps: 15
weight_decay: 0.01
xformers_attention: true
```
</details><br>
# e083b4d3-573f-41e9-8468-fbc799e27adf
This model is a fine-tuned version of [teknium/OpenHermes-2.5-Mistral-7B](https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 1.0812
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 15
- training_steps: 350
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.9191 | 0.1670 | 350 | 1.0812 |
### Framework versions
- PEFT 0.13.2
- Transformers 4.46.0
- Pytorch 2.5.0+cu124
- Datasets 3.0.1
- Tokenizers 0.20.1 |
jackcole/gemma-multimodal-test | jackcole | "2025-05-06T21:34:25Z" | 0 | 0 | transformers | [
"transformers",
"tensorboard",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"base_model:google/gemma-3-4b-pt",
"base_model:finetune:google/gemma-3-4b-pt",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T20:35:36Z" | ---
base_model: google/gemma-3-4b-pt
library_name: transformers
model_name: gemma-multimodal-test
tags:
- generated_from_trainer
- trl
- sft
licence: license
---
# Model Card for gemma-multimodal-test
This model is a fine-tuned version of [google/gemma-3-4b-pt](https://huggingface.co/google/gemma-3-4b-pt).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="jackcole/gemma-multimodal-test", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.15.2
- Transformers: 4.51.3
- Pytorch: 2.6.0+cu124
- Datasets: 3.3.2
- Tokenizers: 0.21.1
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
``` |
dimasik2987/b1dfa283-5e16-4518-b3d7-9d2fde67baa0 | dimasik2987 | "2025-05-06T21:33:56Z" | 0 | 0 | peft | [
"peft",
"safetensors",
"qwen2",
"axolotl",
"generated_from_trainer",
"base_model:Qwen/Qwen2.5-0.5B-Instruct",
"base_model:adapter:Qwen/Qwen2.5-0.5B-Instruct",
"license:apache-2.0",
"4-bit",
"bitsandbytes",
"region:us"
] | null | "2025-05-06T21:20:01Z" | ---
library_name: peft
license: apache-2.0
base_model: Qwen/Qwen2.5-0.5B-Instruct
tags:
- axolotl
- generated_from_trainer
model-index:
- name: b1dfa283-5e16-4518-b3d7-9d2fde67baa0
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.4.1`
```yaml
absolute_data_files: false
adapter: lora
base_model: Qwen/Qwen2.5-0.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: /workspace/axolotl
datasets:
- data_files:
- 122de0a47a4e3406_train_data.json
ds_type: json
format: custom
path: /workspace/input_data/122de0a47a4e3406_train_data.json
type:
field_instruction: premise
field_output: hypothesis
format: '{instruction}'
no_input_format: '{instruction}'
system_format: '{system}'
system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 1
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_clipping: 0.55
group_by_length: false
hub_model_id: dimasik2987/b1dfa283-5e16-4518-b3d7-9d2fde67baa0
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 1.0e-06
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 128
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lr_scheduler: cosine
max_steps: 400
micro_batch_size: 8
mixed_precision: bf16
mlflow_experiment_name: /tmp/122de0a47a4e3406_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 1
sequence_len: 2048
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 493d3ac2-1474-46be-82a1-2a8cda6e563b
wandb_project: s56-28
wandb_run: your_name
wandb_runid: 493d3ac2-1474-46be-82a1-2a8cda6e563b
warmup_steps: 20
weight_decay: 0.01
xformers_attention: true
```
</details><br>
# b1dfa283-5e16-4518-b3d7-9d2fde67baa0
This model is a fine-tuned version of [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 3.4379
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-06
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- total_train_batch_size: 16
- total_eval_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 20
- training_steps: 400
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 3.7336 | 0.0400 | 400 | 3.4379 |
### Framework versions
- PEFT 0.13.2
- Transformers 4.46.0
- Pytorch 2.5.0+cu124
- Datasets 3.0.1
- Tokenizers 0.20.1 |
buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora | buelfhood | "2025-05-06T21:31:16Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T21:31:14Z" | ---
tags:
- roberta
- adapter-transformers
---
# Adapter `buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora` for microsoft/graphcodebert-base
An [adapter](https://adapterhub.ml) for the `microsoft/graphcodebert-base` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("microsoft/graphcodebert-base")
adapter_name = model.load_adapter("buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0003_l512_s42_ppy_f_beta_score_lora", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
mradermacher/Claria-14b-i1-GGUF | mradermacher | "2025-05-06T21:31:03Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"text-generation-inference",
"unsloth",
"qwen3",
"trl",
"sft",
"en",
"base_model:drwlf/Claria-14b",
"base_model:quantized:drwlf/Claria-14b",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | "2025-05-06T08:16:06Z" | ---
base_model: drwlf/Claria-14b
language:
- en
library_name: transformers
license: apache-2.0
quantized_by: mradermacher
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
- trl
- sft
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: nicoboss -->
weighted/imatrix quants of https://huggingface.co/drwlf/Claria-14b
<!-- provided-files -->
static quants are available at https://huggingface.co/mradermacher/Claria-14b-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ1_S.gguf) | i1-IQ1_S | 3.7 | for the desperate |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ1_M.gguf) | i1-IQ1_M | 3.9 | mostly desperate |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 4.4 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ2_XS.gguf) | i1-IQ2_XS | 4.8 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ2_S.gguf) | i1-IQ2_S | 5.1 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ2_M.gguf) | i1-IQ2_M | 5.4 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q2_K_S.gguf) | i1-Q2_K_S | 5.5 | very low quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q2_K.gguf) | i1-Q2_K | 5.9 | IQ3_XXS probably better |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 6.0 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ3_XS.gguf) | i1-IQ3_XS | 6.5 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q3_K_S.gguf) | i1-Q3_K_S | 6.8 | IQ3_XS probably better |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ3_S.gguf) | i1-IQ3_S | 6.8 | beats Q3_K* |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ3_M.gguf) | i1-IQ3_M | 7.0 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q3_K_M.gguf) | i1-Q3_K_M | 7.4 | IQ3_S probably better |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q3_K_L.gguf) | i1-Q3_K_L | 8.0 | IQ3_M probably better |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ4_XS.gguf) | i1-IQ4_XS | 8.2 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-IQ4_NL.gguf) | i1-IQ4_NL | 8.6 | prefer IQ4_XS |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q4_0.gguf) | i1-Q4_0 | 8.6 | fast, low quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q4_K_S.gguf) | i1-Q4_K_S | 8.7 | optimal size/speed/quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q4_K_M.gguf) | i1-Q4_K_M | 9.1 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q4_1.gguf) | i1-Q4_1 | 9.5 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q5_K_S.gguf) | i1-Q5_K_S | 10.4 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q5_K_M.gguf) | i1-Q5_K_M | 10.6 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-i1-GGUF/resolve/main/Claria-14b.i1-Q6_K.gguf) | i1-Q6_K | 12.2 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
<!-- end -->
|
opennomad/ShellLife | opennomad | "2025-05-06T21:30:15Z" | 27 | 0 | transformers | [
"transformers",
"safetensors",
"gguf",
"mistral",
"text-generation-inference",
"unsloth",
"trl",
"en",
"dataset:opennomad/ShellLife",
"base_model:unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
"base_model:quantized:unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-04-09T17:28:26Z" | ---
base_model: unsloth/mistral-7b-instruct-v0.3-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- mistral
- trl
license: apache-2.0
language:
- en
datasets:
- opennomad/ShellLife
---
# Uploaded model
- **Developed by:** opennomad
- **License:** apache-2.0
- **Finetuned from model :** unsloth/mistral-7b-instruct-v0.3-bnb-4bit
This mistral model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth) |
mradermacher/Claria-14b-GGUF | mradermacher | "2025-05-06T21:29:37Z" | 37 | 1 | transformers | [
"transformers",
"gguf",
"text-generation-inference",
"unsloth",
"qwen3",
"trl",
"sft",
"en",
"base_model:drwlf/Claria-14b",
"base_model:quantized:drwlf/Claria-14b",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-05T02:43:47Z" | ---
base_model: drwlf/Claria-14b
language:
- en
library_name: transformers
license: apache-2.0
quantized_by: mradermacher
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
- trl
- sft
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/drwlf/Claria-14b
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/Claria-14b-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q2_K.gguf) | Q2_K | 5.9 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q3_K_S.gguf) | Q3_K_S | 6.8 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q3_K_M.gguf) | Q3_K_M | 7.4 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q3_K_L.gguf) | Q3_K_L | 8.0 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.IQ4_XS.gguf) | IQ4_XS | 8.3 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q4_K_S.gguf) | Q4_K_S | 8.7 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q4_K_M.gguf) | Q4_K_M | 9.1 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q5_K_S.gguf) | Q5_K_S | 10.4 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q5_K_M.gguf) | Q5_K_M | 10.6 | |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q6_K.gguf) | Q6_K | 12.2 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Claria-14b-GGUF/resolve/main/Claria-14b.Q8_0.gguf) | Q8_0 | 15.8 | fast, best quality |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
Zack-Z/qwen3_4bi_cotsft_rs0_3_5cut_gem3_e2 | Zack-Z | "2025-05-06T21:28:26Z" | 0 | 0 | transformers | [
"transformers",
"qwen3",
"feature-extraction",
"text-generation-inference",
"unsloth",
"en",
"base_model:unsloth/Qwen3-4B",
"base_model:finetune:unsloth/Qwen3-4B",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | "2025-05-06T21:14:45Z" | ---
base_model: unsloth/Qwen3-4B
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** Zack-Z
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Qwen3-4B
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora | buelfhood | "2025-05-06T21:28:01Z" | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"region:us"
] | null | "2025-05-06T21:27:55Z" | ---
tags:
- adapter-transformers
- roberta
---
# Adapter `buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora` for microsoft/graphcodebert-base
An [adapter](https://adapterhub.ml) for the `microsoft/graphcodebert-base` model that was trained on the None dataset and includes a prediction head for classification.
This adapter was created for usage with the **[Adapters](https://github.com/Adapter-Hub/adapters)** library.
## Usage
First, install `adapters`:
```
pip install -U adapters
```
Now, the adapter can be loaded and activated like this:
```python
from adapters import AutoAdapterModel
model = AutoAdapterModel.from_pretrained("microsoft/graphcodebert-base")
adapter_name = model.load_adapter("buelfhood/conplag1_graphcodebert_ep50_bs16_lr0_0001_l512_s42_ppy_f_beta_score_lora", set_active=True)
```
## Architecture & Training
<!-- Add some description here -->
## Evaluation results
<!-- Add some description here -->
## Citation
<!-- Add some description here --> |
mfurkan03/custom-yolo-model | mfurkan03 | "2025-05-06T21:27:41Z" | 0 | 0 | ultralytics | [
"ultralytics",
"safetensors",
"object-detection",
"computer-vision",
"yolov10",
"dataset:detection-datasets/coco",
"arxiv:2405.14458",
"license:agpl-3.0",
"region:us"
] | object-detection | "2025-05-06T18:24:52Z" | ---
license: agpl-3.0
library_name: ultralytics
repo_url: https://github.com/THU-MIG/yolov10
tags:
- object-detection
- computer-vision
- yolov10
datasets:
- detection-datasets/coco
inference: false
---
### Model Description
[YOLOv10: Real-Time End-to-End Object Detection](https://arxiv.org/abs/2405.14458v1)
- arXiv: https://arxiv.org/abs/2405.14458v1
- github: https://github.com/THU-MIG/yolov10
### Installation
```
pip install git+https://github.com/THU-MIG/yolov10.git
```
### Training and validation
```python
from ultralytics import YOLOv10
model = YOLOv10.from_pretrained('jameslahm/yolov10n')
# Training
model.train(...)
# after training, one can push to the hub
model.push_to_hub("your-hf-username/yolov10-finetuned")
# Validation
model.val(...)
```
### Inference
Here's an end-to-end example showcasing inference on a cats image:
```python
from ultralytics import YOLOv10
model = YOLOv10.from_pretrained('jameslahm/yolov10n')
source = 'http://images.cocodataset.org/val2017/000000039769.jpg'
model.predict(source=source, save=True)
```
which shows:
### BibTeX Entry and Citation Info
```
@article{wang2024yolov10,
title={YOLOv10: Real-Time End-to-End Object Detection},
author={Wang, Ao and Chen, Hui and Liu, Lihao and Chen, Kai and Lin, Zijia and Han, Jungong and Ding, Guiguang},
journal={arXiv preprint arXiv:2405.14458},
year={2024}
}
``` |
mradermacher/Xiaolong-Qwen3-0.6B-GGUF | mradermacher | "2025-05-06T21:25:56Z" | 0 | 0 | transformers | [
"transformers",
"gguf",
"orpo",
"uncensored",
"reasoning",
"cot",
"en",
"dataset:nbeerbower/GreatFirewall-DPO",
"dataset:nbeerbower/Schule-DPO",
"dataset:nbeerbower/Purpura-DPO",
"dataset:nbeerbower/Arkhaios-DPO",
"dataset:jondurbin/truthy-dpo-v0.1",
"dataset:antiven0m/physical-reasoning-dpo",
"dataset:flammenai/Date-DPO-NoAsterisks",
"dataset:flammenai/Prude-Phi3-DPO",
"dataset:Atsunori/HelpSteer2-DPO",
"dataset:jondurbin/gutenberg-dpo-v0.1",
"dataset:nbeerbower/gutenberg2-dpo",
"dataset:nbeerbower/gutenberg-moderne-dpo",
"dataset:GeneralReasoning/GeneralThought-430K",
"dataset:nvidia/OpenMathReasoning",
"dataset:nvidia/OpenCodeReasoning",
"base_model:nbeerbower/Xiaolong-Qwen3-0.6B",
"base_model:quantized:nbeerbower/Xiaolong-Qwen3-0.6B",
"license:apache-2.0",
"endpoints_compatible",
"region:us",
"conversational"
] | null | "2025-05-06T21:13:24Z" | ---
base_model: nbeerbower/Xiaolong-Qwen3-0.6B
datasets:
- nbeerbower/GreatFirewall-DPO
- nbeerbower/Schule-DPO
- nbeerbower/Purpura-DPO
- nbeerbower/Arkhaios-DPO
- jondurbin/truthy-dpo-v0.1
- antiven0m/physical-reasoning-dpo
- flammenai/Date-DPO-NoAsterisks
- flammenai/Prude-Phi3-DPO
- Atsunori/HelpSteer2-DPO
- jondurbin/gutenberg-dpo-v0.1
- nbeerbower/gutenberg2-dpo
- nbeerbower/gutenberg-moderne-dpo
- GeneralReasoning/GeneralThought-430K
- nvidia/OpenMathReasoning
- nvidia/OpenCodeReasoning
language:
- en
library_name: transformers
license: apache-2.0
quantized_by: mradermacher
tags:
- orpo
- uncensored
- reasoning
- cot
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
static quants of https://huggingface.co/nbeerbower/Xiaolong-Qwen3-0.6B
<!-- provided-files -->
weighted/imatrix quants are available at https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-i1-GGUF
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q2_K.gguf) | Q2_K | 0.4 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q3_K_S.gguf) | Q3_K_S | 0.4 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q3_K_M.gguf) | Q3_K_M | 0.4 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q3_K_L.gguf) | Q3_K_L | 0.5 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.IQ4_XS.gguf) | IQ4_XS | 0.5 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q4_K_S.gguf) | Q4_K_S | 0.5 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q4_K_M.gguf) | Q4_K_M | 0.5 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q5_K_S.gguf) | Q5_K_S | 0.5 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q5_K_M.gguf) | Q5_K_M | 0.5 | |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q6_K.gguf) | Q6_K | 0.6 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.Q8_0.gguf) | Q8_0 | 0.7 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Xiaolong-Qwen3-0.6B-GGUF/resolve/main/Xiaolong-Qwen3-0.6B.f16.gguf) | f16 | 1.3 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
gradientrouting-spar/qwen_ft_May3_m4_p1_num1 | gradientrouting-spar | "2025-05-06T21:25:35Z" | 0 | 0 | transformers | [
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | "2025-05-06T21:24:48Z" | ---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed] |
Subsets and Splits