asyafiqe/Merak-7B-v2-GGML

MERAK-7B-V2 GGML

readme adapted from TheBloke

These files are GGML format model files for MERAK-7B-V2.

GGML files are for CPU + GPU inference using llama.cpp and libraries and UIs which support this format, such as:

• KoboldCpp, a powerful GGML web UI with full GPU acceleration out of the box. Especially good for story telling.
• LoLLMS Web UI, a great web UI with GPU acceleration via the c_transformers backend.
• LM Studio, a fully featured local GUI. Supports full GPU accel on macOS. Also supports Windows, without GPU accel.
• text-generation-webui, the most popular web UI. Requires extra steps to enable GPU accel via llama.cpp backend.
• ctransformers, a Python library with LangChain support and OpenAI-compatible AI server.
• llama-cpp-python, a Python library with OpenAI-compatible API server.

Compatibility

Original llama.cpp quant methods: q4_0, q4_1, q5_0, q5_1, q8_0

These are guaranteed to be compatible with any UIs, tools and libraries released since late May. They may be phased out soon, as they are largely superseded by the new k-quant methods.

New k-quant methods: q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K

These new quantisation methods are compatible with llama.cpp as of June 6th, commit 2d43387.

They are now also compatible with recent releases of text-generation-webui, KoboldCpp, llama-cpp-python, ctransformers, rustformers and most others. For compatibility with other tools and libraries, please check their documentation.

Explanation of the new k-quant methods

Click to see details

The new methods available are:

• GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
• GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
• GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
• GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
• GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
• GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type.

Refer to the Provided Files table below to see what files use which methods, and how.

Provided files

Name	Quant method	Bits	Use case
Merak-7B-v2.ggmlv3.q2_K.bin	q2_K	2	New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors.
Merak-7B-v2.ggmlv3.q3_K_L.bin	q3_K_L	3	New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
Merak-7B-v2.ggmlv3.q3_K_M.bin	q3_K_M	3	New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
Merak-7B-v2.ggmlv3.q3_K_S.bin	q3_K_S	3	New k-quant method. Uses GGML_TYPE_Q3_K for all tensors
Merak-7B-v2.ggmlv3.q4_0.bin	q4_0	4	Original quant method, 4-bit.
Merak-7B-v2.ggmlv3.q4_1.bin	q4_1	4	Original quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
Merak-7B-v2.ggmlv3.q4_K_M.bin	q4_K_M	4	New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K
Merak-7B-v2.ggmlv3.q4_K_S.bin	q4_K_S	4	New k-quant method. Uses GGML_TYPE_Q4_K for all tensors
Merak-7B-v2.ggmlv3.q5_0.bin	q5_0	5	Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference.
Merak-7B-v2.ggmlv3.q5_1.bin	q5_1	5	Original quant method, 5-bit. Even higher accuracy, resource usage and slower inference.
Merak-7B-v2.ggmlv3.q5_K_M.bin	q5_K_M	5	New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K
Merak-7B-v2.ggmlv3.q5_K_S.bin	q5_K_S	5	New k-quant method. Uses GGML_TYPE_Q5_K for all tensors
Merak-7B-v2.ggmlv3.q6_K.bin	q6_K	6	New k-quant method. Uses GGML_TYPE_Q8_K for all tensors - 6-bit quantization
lMerak-7B-v2.ggmlv3.q8_0.bin	q8_0	8	Original quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users.

How to run in text-generation-webui

Further instructions here: text-generation-webui/docs/llama.cpp-models.md.

Original model card: 6TH PROTOTYPE OF MERAK-7B-V2!

Merak-7B is the Large Language Model of Indonesia Languange

This model is based on Meta Llama-2-7B-Chat-HF and fine tuned by some of Indonesia Wikipedia articles that I cleaned before.

Leveraging QLoRA (QLora: Efficient Finetuning of Quantized LLMs), Merak-7B is able to run with 16 GB VRAM

Licensed under Creative Commons-By Attribution-Share Alike-Non Commercial (CC-BY-SA-NC 4.0) Merak-7B empowers AI enthusiasts, researchers alike.

Big thanks to all my friends and communities that help to build our first model. Feel free, to ask me about the model and please share the news on your social media.

HOW TO USE

Installation

Please make sure you have installed CUDA driver in your system, Python 3.10 and PyTorch 2. Then install this library in terminal

pip install bitsandbytes==0.39.1
pip install transformers==4.31.0
pip install peft==0.4.0
pip install accelerate==0.20.3
pip install einops==0.6.1 scipy sentencepiece datasets

Using BitsandBytes and it run with >= 10 GB VRAM GPU

import torch
from transformers import AutoTokenizer, AutoConfig, AutoModelForCausalLM, BitsAndBytesConfig, LlamaTokenizer
from peft import PeftModel, PeftConfig

model_id = "Ichsan2895/Merak-7B-v2"
config = AutoConfig.from_pretrained(model_id)

BNB_CONFIG = BitsAndBytesConfig(load_in_4bit=True,
                                bnb_4bit_compute_dtype=torch.bfloat16,
                                bnb_4bit_use_double_quant=True,
                                bnb_4bit_quant_type="nf4",
    )

model = AutoModelForCausalLM.from_pretrained(model_id,
                                             quantization_config=BNB_CONFIG,
                                             device_map="auto",
                                             trust_remote_code=True)

tokenizer = LlamaTokenizer.from_pretrained(model_id)

def generate_response(question: str) -> str:
  prompt = f"<|prompt|>{question}\n<|answer|>".strip()

  encoding = tokenizer(prompt, return_tensors='pt').to("cuda")
  with torch.inference_mode():
    outputs = model.generate(input_ids=encoding.input_ids,
                             attention_mask=encoding.attention_mask,
                             eos_token_id=tokenizer.pad_token_id,
                             do_sample=False,
                             num_beams=2,
                             temperature=0.3,
                             repetition_penalty=1.2,
                             max_length=200)

    response = tokenizer.decode(outputs[0], skip_special_tokes=True)

    assistant_start = "<|answer|>"
    response_start = response.find(assistant_start)
return response[response_start + len(assistant_start) :].strip()

prompt = "Siapa penulis naskah proklamasi kemerdekaan Indonesia?"
print(generate_response(prompt))

From my experience, For better answer, please don’t use BitsandBytes 4-bit Quantization, but it using higher VRAM

import torch
from transformers import AutoTokenizer, AutoConfig, AutoModelForCausalLM, BitsAndBytesConfig, LlamaTokenizer
from peft import PeftModel, PeftConfig

model_id = "Ichsan2895/Merak-7B-v2"
config = AutoConfig.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id,
                                             device_map="auto",
                                             trust_remote_code=True)

tokenizer = LlamaTokenizer.from_pretrained(model_id)

def generate_response(question: str) -> str:
  prompt = f"<|prompt|>{question}\n<|answer|>".strip()

  encoding = tokenizer(prompt, return_tensors='pt').to("cuda")
  with torch.inference_mode():
    outputs = model.generate(input_ids=encoding.input_ids,
                             attention_mask=encoding.attention_mask,
                             eos_token_id=tokenizer.pad_token_id,
                             do_sample=False,
                             num_beams=2,
                             temperature=0.3,
                             repetition_penalty=1.2,
                             max_length=200)

    response = tokenizer.decode(outputs[0], skip_special_tokes=True)

    assistant_start = "<|answer|>"
    response_start = response.find(assistant_start)
return response[response_start + len(assistant_start) :].strip()

prompt = "Siapa penulis naskah proklamasi kemerdekaan Indonesia?"
print(generate_response(prompt))

CHANGELOG

v1 = The first Merak-7B model. We selected and cleaned about 200k ID wikipedia articles.
v2 = Finetuned version of first Merak-7B model. We finetuned again with the same ID Wikipedia articles except it changes prompt-style in the questions.

CITATION

@Paper{arXiv,
  author  = {Touvron, et al},
  title   = {Llama 2: Open Foundation and Fine-Tuned Chat Models},
  journal = {arXiv preprint arXiv:2307.09288},
  year    = {2023}
}

@ONLINE{wikidump,
    author = "Wikimedia Foundation",
    title  = "Wikimedia Downloads",
    url    = "https://dumps.wikimedia.org"
}

@inproceedings{wolf-etal-2020-transformers,
    title = "Transformers: State-of-the-Art Natural Language Processing",
    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
    month = oct,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
    pages = "38--45"
}

@article{dettmers2023qlora,
  title   = {QLoRA: Efficient Finetuning of Quantized LLMs},
  author  = {Dettmers, Tim and Pagnoni, Artidoro and Holtzman, Ari and Zettlemoyer, Luke},
  journal = {arXiv preprint arXiv:2305.14314},
  year    = {2023}
}