Previous Releases

For latest (...), use uv pip install vllm --torch-backend auto

Loading releases...

Installation Notes

• Commands use uv for faster installation. Replace uv pip with pip if needed
• Available platforms are fetched from the wheels index. Not all CUDA versions are available for all releases
• vLLM < v0.10.0: Recommended to use Python 3.12 (uv venv --python 3.12)
• vLLM v0.9.x: These versions require transformers < 4.54.0
• vLLM < v0.6.0: These versions have additional dependency issues that require manual fixes
• ROCm: Available from v0.14.0+. vLLM < 0.19.0 uses ROCm 7.0.0, v0.19.0–v0.20.x uses ROCm 7.2.1, v0.21.0–v0.22.x uses ROCm 7.2.2, vLLM ≥ 0.23.0 uses ROCm 7.2.3. Requires Python 3.12 and glibc ≥ 2.35 before v0.21.0, ≥ 2.34 for v0.21.0–v0.27.x, ≥ 2.39 (Ubuntu 24.04+) from v0.28.0
• CPU: Requires glibc ≥ 2.35 (Ubuntu 22.04+, Debian 12+, RHEL 9+)
• For the latest version, use uv pip install vllm --torch-backend auto
• For more installation options, see installation docs