# turboderp-org/exllamav2

A fast inference library for running LLMs locally on modern consumer-class GPUs

Repository: https://github.com/turboderp-org/exllamav2
Canonical: https://ross.abutalabs.com/products/exllamav2
Language: Python
License: MIT
License Family: permissive
Last push: 2026-03-04T13:12:20+00:00

## Health v2 (maintenance only)
Score: 61/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 70, release rhythm 40, longevity 78
- inputs: {"age_days": 1099, "days_push": 182, "days_rel": 416, "gap_med": 19, "n_releases_24m": 12}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4611, forks 341 (observed 2026-08-28T04:08:54.999357+00:00)

## What it is
ExLlamaV2 is a fast Python inference library for running large language models locally on modern consumer GPUs, with support for EXL2/GPTQ quantization, paged attention via Flash Attention, and dynamic batched or streamed generation. The project is now archived, with development continuing in ExLlamaV3.

## Use cases
- run local llms on my gpu
- fast inference for quantized llama models
- serve an openai-compatible local llm api
- generate text from gptq quantized models
- stream llm token generation with asyncio
- batch multiple prompts for gpu inference

## When to choose
- you want maximum token throughput for quantized LLMs on consumer NVIDIA GPUs
- you need a lightweight Python inference engine rather than a full server
- you are pairing it with TabbyAPI for an OpenAI-compatible local endpoint

## When to avoid
- you are starting a new project - use ExLlamaV3 instead since this repo is archived
- you need CPU-only or Apple Silicon-first inference
- you want a turnkey chat UI rather than an inference library

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-inference, machine-learning
- domain: large-language-models, machine-learning, gpu-computing
- platform: python, windows
- tags: quantization, exl2, gptq, local-llm, consumer-gpus, speculative-decoding, archived, gpu, linux, macos

## Member repositories
- turboderp-org/exllamav2 (main) score 61

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:54.999357+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 2026-08-29T18:19:43.673042+00:00, confidence not recorded.
  - readme: https://github.com/turboderp-org/exllamav2 (fetched 2026-08-28T04:08:54.999357+00:00, sha 9cbf6c8bd9d9)
- Data as of 2026-08-30T08:39:29.467469+00:00.