GPTQModel

Production ready LLM model compression/quantization toolkit with accelerated inference support for both cpu/gpu via HF, vLLM, and SGLang.

News

12/19/2024 1.4.5: Windows 11 support added/validated. Ovis VL model support with image dataset calibration. Fixed dynamic loading. Reduced quantization vram usage.
12/15/2024 1.4.2: MacOS gpu (Metal) and cpu (M+) support added/validated for inference and quantization. Cohere 2 model support added.
12/13/2024 1.4.1: Added Qwen2-VL model support. mse quantization control exposed in QuantizeConfig. Monkey patch patch_vllm() and patch_hf() api added to allow Transformers/Optimum/PEFT and vLLM to correctly loaded GPTQModel quantized models while upstream PRs are in pending status.
12/10/2024 1.4.0 EvalPlus harness integration merged upstream. We now support both lm-eval and EvalPlus. Added pure torch Torch kernel. Refactored Cuda kernel to be DynamicCuda kernel. Triton kernel now auto-padded for max model support. Dynamic quantization now supports both positive +::default, and -: negative matching which allows matched modules to be skipped entirely for quantization. Fixed auto-Marlin kerenl selection. Added auto-kernel fallback for unsupported kernel/module pairs. Lots of internal refractor and cleanup in-preparation for transformers/optimum/peft upstream PR merge. Deprecated the saving of Marlin weight format since Marlin supports auto conversion of gptq format to Marlin during runtime.
11/29/2024 1.3.1 Olmo2 model support. Intel XPU acceleration via IPEX. Model sharding Transformer compat fix due to api deprecation in HF. Removed triton dependency. Triton kernel now optionally dependent on triton pkg.
11/26/2024 1.3.0 Zero-Day Hymba model support. Removed tqdm and rogue dependency.
11/24/2024 1.2.3 HF GLM model support. ClearML logging integration. Use device-smi and replace gputil + psutil depends. Fixed model unit tests.

Archived News:

* 11/11/2024 🚀 [1.2.1](https://github.com/ModelCloud/GPTQModel/releases/tag/v1.2.1) Meta MobileLLM model support added. `lm-eval[gptqmodel]` integration merged upstream. Intel/IPEX cpu inference merged replacing QBits (deprecated). Auto-fix/patch ChatGLM-3/GLM-4 compat with latest transformers. New `.load()` and `.save()` api.

Why should you use GPTQModel?

GPTQModel started out as a major refractor (fork) of AutoGTQP but has now morphed into a full-stand-in replacement with cleaner api, up-to-date model support, faster inference, faster quantization, higher quality quants and a pledge that ModelCloud, together with the open-source ML community, will take every effort to bring the library up-to-date with latest advancements and model support.

Why GPTQ and not other low-bit quantizers?

Public tests/papers and ModelCloud's internal tests have shown that GPTQ is on-par and/or exceeds other 4bit quantization methods in terms of both quality recovery and production-level inference speed for token latency and rps. GPTQ has the optimal blend of quality and inference speed you would want to use in a real-world production system.

Features

🚀 Extensive model support for: Ovis VL, Llama 1-3.3, Qwen2-VL, Olmo2, Hymba, GLM, IBM Granite, Llama 3.2 Vision, MiniCPM3, GRIN-Moe, Phi 1-4, EXAONE 3.0, InternLM 2.5, Gemma 2, DeepSeek-V2, DeepSeek-V2-Lite, ChatGLM, MiniCPM, Qwen2MoE, DBRX.
✨ Linux, MacOS, Windows platform quantization and accelerated inference support.
💯 100% CI unit-test coverage for all supported models and kernels including post-quantization quality regression.
✨ Dynamic mixed quantization control on a per-module basis. Each layer/module can have a unique quantization config or be excluded from quantization all together.
🚀 vLLM and SGLang inference integration for quantized model where format = FORMAT.GPTQ
🚀 Intel/IPEX hardware accelerated quantization/inference for CPU [avx, amx, xmx] and Intel GPU [Arc + Datacenter Max].
🚀 Microsoft/BITBLAS format + dynamically compiled inference.
✨ Intel/AutoRound alternative gptq-inference compatible quantization method.
✨ Asymmetric Sym=False support.
✨ lm_head module quant inference support for further VRAM reduction (auto-round only).
🚀 Faster quantization: More than 50% faster for TinyLlama + 4090 with batching and large calibration dataset.
✨ Model weights sharding support with optional hash check of model weights on load.
🚀 40% faster packing stage in quantization (Llama 3.1 8B). 50% faster PPL calculations (OPT).

Quality: GPTQModel 4bit can match BF16:

🤗 ModelCloud quantized ultra-high recovery vortex-series models on HF

Model Support: 🚀 (GPTQModel)

Model
Baichuan	✅	Falcon	✅	Llama 1-3.3	✅	OLMo2	🚀	Yi	✅
Bloom	✅	Gemma 2	🚀	Llama 3.2 VL	🚀	Ovis 1.6	🚀
ChatGLM	🚀	GPTBigCod	✅	LongLLaMA	✅	Phi 1-4	🚀
CodeGen	✅	GPTNeoX	✅	MiniCPM3	✅	Qwen	✅
Cohere 1-2	✅	GPT-2	✅	Mistral	✅	Qwen2 MoE	🚀
DBRX Converted	🚀	GPT-J	✅	Mixtral	✅	Qwen2 VL	🚀
Deci	✅	Granite	🚀	MobileLLM	🚀	RefinedWeb	✅
DeepSeek-V2	🚀	GRIN-MoE	🚀	MOSS	✅	StableLM	✅
DeepSeek-V2-Lite	🚀	Hymba	🚀	MPT	✅	StarCoder2	✅
EXAONE 3.0	🚀	InternLM 1/2.5	🚀	OPT	✅	XVERSE	✅

Platform and HW Support

GPTQModel is validated for Linux, MacOS, and Windows 11:

Platform	Device		Optimized Arch	Kernels
Linux	Nvidia GPU	✅	Ampere or Higher	Marlin, Exllama V2, Exallma V1, Triton, DyanamicCuda, Torch
Linux	Intel/AMD CPU	✅	`avx512` or `amx`	IPEX, Torch
Linux	Intel XPU	✅	Intel Arc + Datacenter Max	IPEX, Torch
MacOS	GPU (Metal) / CPU	✅	M1+	Torch
Windows 11	GPU (Nvidia) / CPU	✅	Nvidia	DynamicCuda, Torch

Install

PIP/UV

# You can install optional modules like autoround, ipex, vllm, sglang, bitblas, and ipex.
# Example: pip install -v --no-build-isolation gptqmodel[vllm,sglang,bitblas,ipex,auto_round]
pip install -v gptqmodel --no-build-isolation
uv pip install -v gptqmodel --no-build-isolation

Install from source

# clone repo
git clone https://github.com/ModelCloud/GPTQModel.git && cd GPTQModel

# pip: compile and install
# You can install optional modules like autoround, ipex, vllm, sglang, bitblas, and ipex.
# Example: pip install -v --no-build-isolation .[vllm,sglang,bitblas,ipex,auto_round]
pip install -v . --no-build-isolation

Quantization and Inference

Below is a basic sample using GPTQModel to quantize a llm model and perform post-quantization inference:

from datasets import load_dataset
from transformers import AutoTokenizer
from gptqmodel import GPTQModel, QuantizeConfig

model_id = "meta-llama/Llama-3.2-1B-Instruct"
quant_path = "Llama-3.2-1B-Instruct-gptqmodel-4bit"

tokenizer = AutoTokenizer.from_pretrained(model_id)

calibration_dataset = [
  tokenizer(example["text"])
  for example in load_dataset(
    "allenai/c4",
    data_files="en/c4-train.00001-of-01024.json.gz",
    split="train"
  ).select(range(1024))
]

quant_config = QuantizeConfig(bits=4, group_size=128)

model = GPTQModel.load(model_id, quant_config)

model.quantize(calibration_dataset)

model.save(quant_path)

model = GPTQModel.load(quant_path)

result = model.generate(
  **tokenizer(
      "Uncovering deep insights begins with", return_tensors="pt"
  ).to(model.device)
)[0]

For more advanced features of model quantization, please reference to this script

How to Add Support for a New Model

Read the gptqmodel/models/llama.py code which explains in detail via comments how the model support is defined. Use it as guide to PR for to new models. Most models follow the same pattern.

Evaluation and Quality Benchmarks

GPTQModel inference is integrated into both lm-eval and evalplus
We highly recommend avoid using ppl and use lm-eval/evalplus to validate post-quantization model quality. ppl should only be used for regression tests and is not a good indicator of model output quality.

# gptqmodel is integrated into lm-eval >= v0.4.6
pip install lm-eval>=0.4.6

# gptqmodel is integrated into evalplus[main]
pip install -U "evalplus @ git+https://github.com/evalplus/evalplus"

Below is a basic sample using GPTQModel.eval API

from gptqmodel import GPTQModel
from gptqmodel.utils import EVAL

model_id = "ModelCloud/Llama-3.2-1B-Instruct-gptqmodel-4bit-vortex-v1"

# Use `lm-eval` as framework to evaluate the model
lm_eval_results = GPTQModel.eval(model_id, framework=EVAL.LM_EVAL, tasks=[EVAL.LM_EVAL.ARC_CHALLENGE], output_file='lm-eval_result.json')

# Use `evalplus` as framework to evaluate the model
evalplus_results = GPTQModel.eval(model_id, framework=EVAL.EVALPLUS, tasks=[EVAL.EVALPLUS.HUMAN], output_file='evalplus_result.json')

Citation

@misc{gptqmodel,
    author = {ModelCloud.ai and [email protected]},
    title = {GPTQModel},
    year = {2024},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/modelcloud/gptqmodel}},
    note = {Contact: [email protected]}
}

@article{frantar-gptq,
  title={{GPTQ}: Accurate Post-training Compression for Generative Pretrained Transformers}, 
  author={Elias Frantar and Saleh Ashkboos and Torsten Hoefler and Dan Alistarh},
  year={2022},
  journal={arXiv preprint arXiv:2210.17323}
}

@article{frantar2024marlin,
  title={MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models},
  author={Frantar, Elias and Castro, Roberto L and Chen, Jiale and Hoefler, Torsten and Alistarh, Dan},
  journal={arXiv preprint arXiv:2408.11743},
  year={2024}
}

Oven

gptqmodel1.4.5

Package Downloads

Authors

Project URLs

Requires Python

Dependencies