# turboderp-org/exllamav3

An optimized quantization and inference library for running LLMs locally on modern consumer-class GPUs

Repository: https://github.com/turboderp-org/exllamav3
Canonical: https://ross.abutalabs.com/products/exllamav3
Language: Python
License: MIT
License Family: permissive
Last push: 2026-09-02T16:17:57+00:00

## Health v2 (maintenance only)
Score: 87/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 100, longevity 36
- inputs: {"age_days": 514, "days_push": 0, "days_rel": 1, "gap_med": 5.0, "n_releases_24m": 55}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1274, forks 147 (observed 2026-09-03T02:15:20.015388+00:00)

## What it is
ExLlamaV3 is a Python library for fast quantization and inference of large language models on consumer-class GPUs, featuring the EXL3 quantization format based on QTIP. It supports tensor-parallel and expert-parallel inference, speculative decoding, multimodal models, and ships with an OpenAI-compatible server via TabbyAPI.

## Use cases
- run local llms on my gpu
- quantize llama models to run on consumer hardware
- serve an openai-compatible api for local models
- run large models on a single consumer gpu with quantization
- tensor-parallel inference across multiple gpus
- run qwen or mistral models locally with low vram

## When to choose
- you want maximum inference speed for LLMs on consumer NVIDIA/AMD GPUs
- you need aggressive 2-8 bit quantization with minimal quality loss
- you want an OpenAI-compatible local server via TabbyAPI
- you need tensor-parallel or expert-parallel setups on consumer hardware

## When to avoid
- you need CPU-only inference
- you need production multi-node serving at scale
- you require model architectures not in the supported list
- you want a managed cloud inference solution

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, gpu-computing, http-server, machine-learning
- domain: large-language-models, machine-learning, developer-tools
- platform: python, windows, cli
- tags: quantization, exl3, local-llm, tensor-parallel, speculative-decoding, consumer-gpus, tabbyapi, gpu, linux

## Member repositories
- turboderp-org/exllamav3 (main) score 87

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:20.015388+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 2026-08-30T06:28:44.402887+00:00, confidence not recorded.
  - readme: https://github.com/turboderp-org/exllamav3 (fetched 2026-09-03T02:15:20.015388+00:00, sha 2fa8c685db8b)
  - registry_pypi: https://pypi.org/pypi/exllamav3/json (fetched 2026-08-29T12:34:46.467649+00:00, sha c9ccf2572604)
- Data as of 2026-08-30T08:39:29.467469+00:00.