# vllm-project/vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

Repository: https://github.com/vllm-project/vllm
Canonical: https://ross.abutalabs.com/products/vllm
Homepage: https://vllm.ai
Language: Python
License: Apache-2.0
License Family: permissive
Topics: gpt, llm, pytorch, model-serving, transformer, llm-serving, inference, llama, amd, cuda, tpu, deepseek, qwen, blackwell, deepseek-v3, gpt-oss, kimi, moe, openai, qwen3
Last push: 2026-08-26T23:19:49+00:00

## Health v2 (maintenance only)
Score: 93/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 87, longevity 92
- inputs: {"age_days": 1301, "days_push": 7, "days_rel": 7, "gap_med": 9.5, "n_releases_24m": 61}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 90149, forks 21253 (observed 2026-08-28T04:12:23.259628+00:00)

## What it is
vLLM is a high-throughput, memory-efficient inference and serving engine for large language models, featuring PagedAttention, continuous batching, and broad quantization support. It exposes an OpenAI-compatible API server and runs on NVIDIA, AMD, Intel GPUs, TPUs, and CPUs.

## Use cases
- serve an LLM behind an OpenAI-compatible API
- run high-throughput batch inference on GPUs
- deploy quantized models like GPTQ or AWQ
- serve MoE models like DeepSeek or Qwen3
- generate structured outputs from an LLM
- run multi-LoRA serving for many adapters
- scale LLM inference across multiple GPUs with tensor parallelism

## When to choose
- you need maximum serving throughput and GPU utilization
- you want a drop-in OpenAI-compatible endpoint for self-hosted models
- you need broad hardware support including AMD GPUs and TPUs
- you serve many users concurrently with continuous batching

## When to avoid
- you only need to run a small model locally on a laptop
- you need training or fine-tuning rather than inference
- you want a lightweight client library rather than a serving engine

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, http-server, gpu-computing, machine-learning
- domain: large-language-models, machine-learning, artificial-intelligence, backend
- platform: python, cloud
- tags: paged-attention, openai-compatible-api, model-serving, continuous-batching, quantization, speculative-decoding, moe, distributed-inference, linux, gpu, docker

## Member repositories
- vllm-project/vllm (main) score 93

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:12:23.259628+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 2026-08-29T16:13:51.430384+00:00, confidence not recorded.
  - readme: https://github.com/vllm-project/vllm (fetched 2026-08-28T04:12:23.259628+00:00, sha 292534695eff)
  - homepage: https://vllm.ai (fetched 2026-08-28T17:36:37.632137+00:00, sha f3b9661bb5c3)
  - site_page: https://docs.vllm.ai (fetched 2026-08-28T17:36:37.640659+00:00, sha 97bc8b3408a0)
  - site_page: https://docs.vllm.ai/en/latest/getting_started/installation/gpu/index.html (fetched 2026-08-28T17:36:37.644009+00:00, sha c6034c3eb81f)
  - site_page: https://docs.vllm.ai/en/latest/getting_started/installation/gpu/ (fetched 2026-08-28T17:36:37.645532+00:00, sha c6034c3eb81f)
  - site_page: https://docs.vllm.ai/en/latest/getting_started/installation/cpu/ (fetched 2026-08-28T17:36:37.647009+00:00, sha cd8cb3cec937)
  - registry_pypi: https://pypi.org/pypi/vllm/json (fetched 2026-08-28T17:36:37.648462+00:00, sha 1172693245d4)
  - site_page: https://vllm.ai/releases (fetched 2026-08-28T17:36:37.642380+00:00, sha 1250b9c49cad)
- Data as of 2026-08-30T08:39:29.467469+00:00.