[CI/Build][NVIDIA] Build NIXL with NIXL EP from source in Rubin image (#60090)

Signed-off-by: Artem Perevedentsev <aperevedents@nvidia.com>
This commit is contained in:
Artem Perevedentsev
2026-10-08 02:44:17 +04:00
committed by GitHub
parent ef63c23d35
commit 4ac5a36a8e
6 changed files with 679 additions and 0 deletions
@@ -24,6 +24,11 @@ The `extensions-build` stage can also produce an optional source-built Triton
wheel. `vllm-openai-base` installs that wheel after its other Python
dependencies so dependency resolution cannot restore an older Triton version.
With `BUILD_NIXL=true`, `extensions-build` also builds NIXL, including its NIXL
EP extension for the installed PyTorch, with `tools/build_nixl_with_torch215_for_rubin.py`.
`vllm-openai-base` then replaces the NIXL packages installed from the KV-connector
requirements with these wheels.
> <figure markdown="span">
> ![](../../assets/contributing/dockerfile-stages-dependency.png){ align="center" alt="query" width="100%" }
> </figure>
@@ -439,6 +439,11 @@ BuildKit does not automatically invalidate cached layers when a mutable Git
ref changes. Use `--no-cache-filter extensions-build` to refresh an empty,
branch, or tag revision.
Set `BUILD_NIXL=true` to build NIXL from source. NIXL's release wheels do not
include the NIXL EP extension for the PyTorch nightly used by the Rubin build.
The build uses the NIXL version pinned in `requirements/kv_connectors.txt` and
replaces the NIXL packages installed from the KV-connector requirements.
For `FINAL_BASE_IMAGE`, use the public, multi-arch
`nvidia/cuda:13.4.1-base-ubuntu24.04` image. Set `NCCL_VERSION` to 2.32.3 or
newer, the first NCCL release with Rubin (SM107) support.