# Vahe1994/AQLM

Official Pytorch repository for Extreme Compression of Large Language Models via Additive Quantization https://arxiv.org/pdf/2401.06118.pdf and PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression https://arxiv.org/abs/2405.14852

Repository: https://github.com/Vahe1994/AQLM
Canonical: https://ross.abutalabs.com/products/aqlm
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-02-26T13:58:35+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 69, release rhythm 35, longevity 68
- inputs: {"age_days": 964, "days_push": 188, "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 1329, forks 194 (observed 2026-08-28T04:04:23.801600+00:00)

## What it is
Official PyTorch implementation of AQLM, an extreme LLM compression method via additive quantization, extended with PV-Tuning for finetuning quantized models. It includes a pip-installable inference library for running prequantized models and notebooks for quantization, finetuning, and vLLM serving.

## Use cases
- compress large language models to 1-2 bits
- run AQLM-quantized LLMs for inference
- finetune quantized LLMs with PV-tuning
- serve quantized models with vLLM
- quantize my own LLM with codebooks
- run 1-bit LLM on GPU or CPU

## When to choose
- you need extreme (sub-2-bit) LLM compression with good accuracy
- you want to run or finetune AQLM/PV-tuned prequantized models from Hugging Face
- you're reproducing the AQLM or PV-Tuning research papers

## When to avoid
- you only need standard 4/8-bit quantization where simpler tools like GPTQ or bitsandbytes suffice
- you need production serving without GPU-specific kernels
- you're not working with transformer LLMs

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, llm-training, machine-learning, deep-learning
- domain: large-language-models, machine-learning, deep-learning
- platform: python
- tags: quantization, model-compression, additive-quantization, pv-tuning, pytorch, inference, vllm, research, gpu, linux

## Member repositories
- Vahe1994/AQLM (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:23.801600+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 2026-08-30T04:45:38.001129+00:00, confidence not recorded.
  - readme: https://github.com/Vahe1994/AQLM (fetched 2026-08-28T04:04:23.801600+00:00, sha bfe33666487f)
  - registry_pypi: https://pypi.org/pypi/aqlm/json (fetched 2026-08-29T12:04:41.933397+00:00, sha 4e1249cf54c9)
- Data as of 2026-08-30T08:39:29.467469+00:00.