A high-throughput and memory-efficient inference and serving engine for LLMs
Project Links
Meta
Author: vLLM Team
Requires Python: <3.15,>=3.10
Classifiers
Programming Language
- Python :: 3.10
- Python :: 3.11
- Python :: 3.12
- Python :: 3.13
- Python :: 3.14
Intended Audience
- Developers
- Information Technology
- Science/Research
Topic
- Scientific/Engineering :: Artificial Intelligence
- Scientific/Engineering :: Information Analysis
Easy, fast, and cheap LLM serving for everyone
| Documentation | Blog | Paper | Twitter/X | User Forum | Developer Slack |
🔥 We have built a vLLM website to help you get started with vLLM. Please visit vllm.ai to learn more. For events, please visit vllm.ai/events to join us.
About
vLLM is a fast and easy-to-use library for LLM inference and serving.
Originally developed in the Sky Computing Lab at UC Berkeley, vLLM has grown into one of the most active open-source AI projects built and maintained by a diverse community of many dozens of academic institutions and companies from over 2000 contributors.
vLLM is fast with:
- State-of-the-art serving throughput
- Efficient management of attention key and value memory with PagedAttention
- Continuous batching of incoming requests, chunked prefill, prefix caching
- Fast and flexible model execution with piecewise and full CUDA/HIP graphs
- Quantization: FP8, MXFP8/MXFP4, NVFP4, INT8, INT4, GPTQ/AWQ, GGUF, compressed-tensors, ModelOpt, TorchAO, and more
- Optimized attention kernels including FlashAttention, FlashInfer, TRTLLM-GEN, FlashMLA, and Triton
- Optimized GEMM/MoE kernels for various precisions using CUTLASS, TRTLLM-GEN, CuTeDSL
- Speculative decoding including n-gram, suffix, EAGLE, DFlash
- Automatic kernel generation and graph-level transformations using torch.compile
- Disaggregated prefill, decode, and encode
vLLM is flexible and easy to use with:
- Seamless integration with popular Hugging Face models
- High-throughput serving with various decoding algorithms, including parallel sampling, beam search, and more
- Tensor, pipeline, data, expert, and context parallelism for distributed inference
- Streaming outputs
- Generation of structured outputs using xgrammar or guidance
- Tool calling and reasoning parsers
- OpenAI-compatible API server, plus Anthropic Messages API and gRPC support
- Efficient multi-LoRA support for dense and MoE layers
- Support for NVIDIA GPUs, AMD GPUs, and x86/ARM/PowerPC CPUs. Additionally, diverse hardware plugins such as Google TPUs, Intel Gaudi, IBM Spyre, Huawei Ascend, Rebellions NPU, Apple Silicon, MetaX GPU, and more.
vLLM seamlessly supports 200+ model architectures on Hugging Face, including:
- Decoder-only LLMs (e.g., Llama, Qwen, Gemma)
- Mixture-of-Expert LLMs (e.g., Mixtral, DeepSeek-V3, Qwen-MoE, GPT-OSS)
- Hybrid attention and state-space models (e.g., Mamba, Qwen3.5)
- Multi-modal models (e.g., LLaVA, Qwen-VL, Pixtral)
- Embedding and retrieval models (e.g., E5-Mistral, GTE, ColBERT)
- Reward and classification models (e.g., Qwen-Math)
Find the full list of supported models here.
Getting Started
Install vLLM with uv (recommended) or pip:
uv pip install vllm
Or build from source for development.
Visit our documentation to learn more.
Contributing
We welcome and value any contributions and collaborations. Please check out Contributing to vLLM for how to get involved.
Citation
If you use vLLM for your research, please cite our paper:
@inproceedings{kwon2023efficient,
title={Efficient Memory Management for Large Language Model Serving with PagedAttention},
author={Woosuk Kwon and Zhuohan Li and Siyuan Zhuang and Ying Sheng and Lianmin Zheng and Cody Hao Yu and Joseph E. Gonzalez and Hao Zhang and Ion Stoica},
booktitle={Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles},
year={2023}
}
Contact Us
- For technical questions and feature requests, please use GitHub Issues
- For discussing with fellow users, please use the vLLM Forum
- For coordinating contributions and development, please use Slack
- For security disclosures, please use GitHub's Security Advisories feature
- For collaborations and partnerships, please contact us at collaboration@vllm.ai
Media Kit
- If you wish to use vLLM's logo, please refer to our media kit repo
0.21.0
May 15, 2026
0.20.2
May 10, 2026
0.20.1
May 03, 2026
0.20.0
Apr 27, 2026
0.19.1
Apr 18, 2026
0.19.0
Apr 03, 2026
0.18.1
Mar 31, 2026
0.18.0
Mar 20, 2026
0.17.1
Mar 11, 2026
0.17.0
Mar 07, 2026
0.16.0
Feb 26, 2026
0.15.1
Feb 05, 2026
0.15.0
Jan 29, 2026
0.14.1
Jan 24, 2026
0.14.0
Jan 20, 2026
0.13.0
Dec 19, 2025
0.12.0
Dec 03, 2025
0.11.2
Nov 20, 2025
0.11.1
Nov 19, 2025
0.11.0
Oct 04, 2025
0.10.2
Sep 13, 2025
0.10.1.1
Aug 20, 2025
0.10.1
Aug 19, 2025
0.10.0
Jul 25, 2025
0.9.2
Jul 08, 2025
0.9.1
Jun 10, 2025
0.9.0.1
May 30, 2025
0.9.0
May 28, 2025
0.8.5.post1
May 02, 2025
0.8.5
Apr 28, 2025
0.8.4
Apr 15, 2025
0.8.3
Apr 06, 2025
0.8.2
Mar 25, 2025
0.8.1
Mar 19, 2025
0.8.0
Mar 18, 2025
0.7.3
Feb 20, 2025
0.7.2
Feb 06, 2025
0.7.1
Feb 01, 2025
0.7.0
Jan 27, 2025
0.6.6.post1
Dec 27, 2024
0.6.6
Dec 27, 2024
0.6.5
Dec 18, 2024
0.6.4.post1
Nov 15, 2024
0.6.4
Nov 15, 2024
0.6.3.post1
Oct 17, 2024
0.6.3
Oct 14, 2024
0.6.2
Sep 25, 2024
0.6.1.post2
Sep 13, 2024
0.6.1.post1
Sep 13, 2024
0.6.1
Sep 11, 2024
0.6.0
Sep 05, 2024
0.5.5
Aug 23, 2024
0.5.4
Aug 05, 2024
0.5.3.post1
Jul 23, 2024
0.5.3
Jul 23, 2024
0.5.2
Jul 15, 2024
0.5.1
Jul 06, 2024
0.5.0.post1
Jun 14, 2024
0.5.0
Jun 11, 2024
0.4.3
Jun 01, 2024
0.4.2
May 05, 2024
0.4.1
Apr 24, 2024
0.4.0.post1
Apr 03, 2024
0.4.0
Mar 31, 2024
0.3.3
Mar 01, 2024
0.3.2
Feb 21, 2024
0.3.1
Feb 17, 2024
0.3.0
Jan 31, 2024
0.2.7
Jan 04, 2024
0.2.6
Dec 17, 2023
0.2.5
Dec 14, 2023
0.2.4
Dec 11, 2023
0.2.3
Dec 03, 2023
0.2.2
Nov 19, 2023
0.2.1.post1
Oct 17, 2023
0.2.1
Oct 16, 2023
0.2.0
Sep 28, 2023
0.1.7
Sep 11, 2023
0.1.6
Sep 08, 2023
0.1.5
Sep 08, 2023
0.1.4
Aug 25, 2023
0.1.3
Aug 02, 2023
0.1.2
Jul 05, 2023
0.1.1
Jun 22, 2023
0.1.0
Jun 20, 2023
0.0.1
Jun 19, 2023
Wheel compatibility matrix
Files in release
Extras:
Dependencies:
regex
cachetools
psutil
sentencepiece
numpy
requests
(>=2.26.0)
tqdm
blake3
py-cpuinfo
transformers
(!=5.0.*,!=5.1.*,!=5.2.*,!=5.3.*,!=5.4.*,!=5.5.0,>=4.56.0)
tokenizers
(>=0.21.1)
protobuf
(!=6.30.*,!=6.31.*,!=6.32.*,!=6.33.0.*,!=6.33.1.*,!=6.33.2.*,!=6.33.3.*,!=6.33.4.*,>=5.29.6)
fastapi[standard]
(>=0.115.0)
aiohttp
(>=3.13.3)
openai
(>=2.0.0)
pydantic
(>=2.12.0)
prometheus_client
(>=0.18.0)
pillow
prometheus-fastapi-instrumentator
(>=7.0.0)
tiktoken
(>=0.6.0)
lm-format-enforcer
(==0.11.3)
llguidance
or
(<1.4.0,>=1.3.0)
outlines_core
(==0.2.14)
diskcache
(==5.6.3)
lark
(==1.2.2)
xgrammar
or
(<1.0.0,>=0.2.0)
typing_extensions
(>=4.10)
filelock
(>=3.16.1)
partial-json-parser
pyzmq
(>=25.0.0)
msgspec
gguf
(>=0.17.0)
mistral_common[image]
(>=1.11.2)
opencv-python-headless
(>=4.13.0)
pyyaml
six
(>=1.16.0)
setuptools
(<81.0.0,>=77.0.3)
einops
compressed-tensors
(==0.15.0.1)
depyf
(==0.20.0)
cloudpickle
watchfiles
python-json-logger
ninja
pybase64
cbor2
ijson
setproctitle
openai-harmony
(>=0.0.3)
anthropic
(>=0.71.0)
model-hosting-container-standards
(<1.0.0,>=0.1.14)
mcp
opentelemetry-sdk
(>=1.27.0)
opentelemetry-api
(>=1.27.0)
opentelemetry-exporter-otlp
(>=1.27.0)
opentelemetry-semantic-conventions-ai
(>=0.4.1)
numba
(==0.65.0)
torch
(==2.11.0)
torchaudio
(==2.11.0)
torchvision
(==0.26.0)
flashinfer-python
(==0.6.8.post1)
flashinfer-cubin
(==0.6.8.post1)
apache-tvm-ffi
(==0.1.9)
tilelang
(==0.1.9)
nvidia-cudnn-frontend
(<1.19.0,>=1.13.0)
fastsafetensors
(>=0.2.2)
nvidia-cutlass-dsl
(==4.4.2)
quack-kernels
(>=0.3.3)
tokenspeed-mla
(==0.1.2)