Commit
•
9f603c4
1
Parent(s):
5a06031
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
---
|
| 3 |
+
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
+
base_model: gemma-2-Ifable-9B
|
| 6 |
+
library_name: transformers
|
| 7 |
+
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
[](https://hf.co/QuantFactory)
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
# QuantFactory/gemma-2-Ifable-9B-GGUF
|
| 14 |
+
This is quantized version of [ifable/gemma-2-Ifable-9B](https://huggingface.co/ifable/gemma-2-Ifable-9B) created using llama.cpp
|
| 15 |
+
|
| 16 |
+
# Original Model Card
|
| 17 |
+
|
| 18 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 19 |
+
should probably proofread and complete it, then remove this comment. -->
|
| 20 |
+
|
| 21 |
+
# ifable/gemma-2-Ifable-9B
|
| 22 |
+
|
| 23 |
+
## Training and evaluation data
|
| 24 |
+
|
| 25 |
+
- Gutenberg: https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1
|
| 26 |
+
- Carefully curated proprietary creative writing dataset
|
| 27 |
+
|
| 28 |
+
## Training procedure
|
| 29 |
+
|
| 30 |
+
Training method: SimPO (GitHub - princeton-nlp/SimPO: SimPO: Simple Preference Optimization with a Reference-Free Reward)
|
| 31 |
+
|
| 32 |
+
It achieves the following results on the evaluation set:
|
| 33 |
+
- Loss: 1.0163
|
| 34 |
+
- Rewards/chosen: -21.6822
|
| 35 |
+
- Rewards/rejected: -47.8754
|
| 36 |
+
- Rewards/accuracies: 0.9167
|
| 37 |
+
- Rewards/margins: 26.1931
|
| 38 |
+
- Logps/rejected: -4.7875
|
| 39 |
+
- Logps/chosen: -2.1682
|
| 40 |
+
- Logits/rejected: -17.0475
|
| 41 |
+
- Logits/chosen: -12.0041
|
| 42 |
+
|
| 43 |
+
### Training hyperparameters
|
| 44 |
+
|
| 45 |
+
The following hyperparameters were used during training:
|
| 46 |
+
- learning_rate: 8e-07
|
| 47 |
+
- train_batch_size: 1
|
| 48 |
+
- eval_batch_size: 1
|
| 49 |
+
- seed: 42
|
| 50 |
+
- distributed_type: multi-GPU
|
| 51 |
+
- num_devices: 8
|
| 52 |
+
- gradient_accumulation_steps: 16
|
| 53 |
+
- total_train_batch_size: 128
|
| 54 |
+
- total_eval_batch_size: 8
|
| 55 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 56 |
+
- lr_scheduler_type: cosine
|
| 57 |
+
- lr_scheduler_warmup_ratio: 0.1
|
| 58 |
+
- num_epochs: 1.0
|
| 59 |
+
|
| 60 |
+
### Training results
|
| 61 |
+
|
| 62 |
+
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | Sft Loss |
|
| 63 |
+
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------:|
|
| 64 |
+
| 1.4444 | 0.9807 | 35 | 1.0163 | -21.6822 | -47.8754 | 0.9167 | 26.1931 | -4.7875 | -2.1682 | -17.0475 | -12.0041 | 0.0184 |
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
### Framework versions
|
| 68 |
+
|
| 69 |
+
- Transformers 4.43.4
|
| 70 |
+
- Pytorch 2.3.0a0+ebedce2
|
| 71 |
+
- Datasets 2.20.0
|
| 72 |
+
- Tokenizers 0.19.1
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
We are looking for product manager and operations managers to build applications through our model, and also open for business cooperation, and also AI engineer to join us, contact with : [email protected]
|