Part 4 of 5: Build vLLM

2026-07-22 · 1 min read

Upstream vLLM project does not yet offer first party container for inference on Intel cards. So you will have to build your own container with up to date dependencies.

Clone vLLM

git clone git@github.com:vllm-project/vllm.git

Navigate to project directory

cd vllm

Checkout v0.26.0

git checkout origin/releases/v0.26.0

Update XPU container file with most-up-to-date libraries, replace docker/Dockerfile.xpu with following contents Dockerfile.xpu

Build new container:

podman build -f docker/Dockerfile.xpu -t vllm-xpu:v0.26.0 .