# mit-han-lab/llm-awq

[MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

Repository: https://github.com/mit-han-lab/llm-awq
Canonical: https://ross.abutalabs.com/products/llm-awq
Language: Python
License: MIT
License Family: permissive
Last push: 2025-07-17T19:44:51+00:00

## Health v2 (maintenance only)
Score: 44/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 32, release rhythm 35, longevity 85
- inputs: {"age_days": 1190, "days_push": 412, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3621, forks 321 (observed 2026-08-28T04:08:11.857435+00:00)

## What it is
AWQ is a Python library implementing Activation-aware Weight Quantization for compressing large language models to low-bit (INT3/4) weights with efficient CUDA inference kernels. It includes a model zoo of pre-quantized LLMs and TinyChat, a turn-key on-device inference engine for LLMs and vision-language models on edge devices.

## Use cases
- quantize llama models to 4-bit
- run llm inference on jetson edge device
- compress large language models to fit in limited gpu memory
- speed up llm inference with 4-bit weights
- run vision language model on low-power hardware
- quantize deepseek distilled models
- offline on-device chatbot inference

## When to choose
- you need to run LLMs or VLMs on memory-constrained GPUs or edge devices like Jetson
- you want accurate INT4 weight-only quantization with fast CUDA kernels
- you want pre-quantized model checkpoints for popular LLM families

## When to avoid
- you need quantization of non-LLM model types
- you need CPU-only inference without NVIDIA GPUs
- you need training or fine-tuning rather than inference compression

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, machine-learning, gpu-computing, benchmarking
- domain: large-language-models, machine-learning, gpu-computing, developer-tools
- platform: python, cross-platform
- tags: quantization, weight-quantization, int4, model-compression, inference-optimization, edge-inference, cuda-kernels, vision-language-models, gpu, linux

## Member repositories
- mit-han-lab/llm-awq (main) score 44

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:11.857435+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:31:58.067855+00:00, confidence not recorded.
  - readme: https://github.com/mit-han-lab/llm-awq (fetched 2026-08-28T04:08:11.857435+00:00, sha 970c98b60382)
- Data as of 2026-08-30T08:39:29.467469+00:00.