[XPU][CI/Release][3/N] Add xpu wheel release to release pipeline (#52108)

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
This commit is contained in:
Kunshang Ji
2026-08-14 14:00:17 +08:00
committed by GitHub
parent ac7509e2b1
commit 8e6d8e4f6a
6 changed files with 99 additions and 29 deletions
+15 -1
View File
@@ -164,7 +164,21 @@ steps:
env:
DOCKER_BUILDKIT: "1"
- label: "Publish XPU Triton shim index"
- label: "Build wheel - x86_64 - XPU"
depends_on: ~
id: build-wheel-x86-xpu
agents:
queue: cpu_queue_release
commands:
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag vllm-ci:build-image --target vllm-build --progress plain -f docker/Dockerfile.xpu ."
- "mkdir artifacts"
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
- "bash .buildkite/scripts/upload-nightly-wheels.sh"
- 'bash .buildkite/scripts/annotate-build-artifact.sh "$$BUILDKITE_LABEL" "s3://vllm-wheels/$$BUILDKITE_COMMIT/$(cd artifacts/dist && echo *.whl)" release-wheels'
env:
DOCKER_BUILDKIT: "1"
- label: "Publish and stage XPU Triton shim"
key: publish-xpu-triton-shim
depends_on: ~
agents:
@@ -53,11 +53,10 @@ if [[ "${UPDATE_NIGHTLY_INDEX:-1}" == "1" && \
aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "s3://$BUCKET/nightly/"
fi
# detect version from any wheel in the commit directory
# download the first wheel we find to extract version metadata
first_wheel_key=$($PYTHON -c "import json; obj=json.load(open('$obj_json')); print(next((c['Key'] for c in obj.get('Contents', []) if c['Key'].endswith('.whl')), ''))")
# detect version from a vLLM wheel in the commit directory
first_wheel_key=$($PYTHON -c "import json; obj=json.load(open('$obj_json')); print(next((c['Key'] for c in obj.get('Contents', []) if c['Key'].rsplit('/', 1)[-1].startswith('vllm-') and c['Key'].endswith('.whl')), ''))")
if [[ -z "$first_wheel_key" ]]; then
echo "Error: No wheels found in $S3_COMMIT_PREFIX"
echo "Error: No vLLM wheel found in $S3_COMMIT_PREFIX"
exit 1
fi
first_wheel=$(basename "$first_wheel_key")
+7 -3
View File
@@ -91,9 +91,9 @@ def parse_from_filename(file: str) -> WheelFileInfo:
else:
if "+" in version:
version_part, suffix = version.split("+", 1)
# Only treat known patterns as variants (rocmXXX, cuXXX, cpu)
# Only treat known patterns as variants (rocmXXX, cuXXX, cpu, xpu)
# Git hashes and other suffixes are NOT variants
if suffix.startswith(("rocm", "cu", "cpu")):
if suffix.startswith(("rocm", "cu", "cpu", "xpu")):
variant = suffix
version = version_part
# Otherwise keep the full version string (variant stays None)
@@ -429,7 +429,11 @@ if __name__ == "__main__":
if PY_VERSION_RE.match(version):
# upload-wheels.sh ensures no "dev" is in args.version
wheel_files = list(
filter(lambda x: version in x and "dev" not in x, wheel_files)
filter(
lambda x: (version in x and "dev" not in x)
or (x.startswith("triton-") and "+xpu-" in x),
wheel_files,
)
)
print(f"Non-nightly version detected, wheel files used: {wheel_files}")
else:
+11 -3
View File
@@ -11,12 +11,17 @@ readonly WHEEL_SHA256="3c822f73e9870512f59a6ecf5dc305a4bcab11fa623f9ce91011f6043
readonly WHEEL_FILENAME="${WHEEL_URL##*/}"
readonly ENCODED_WHEEL_FILENAME="${WHEEL_FILENAME/+/%2B}"
readonly S3_PREFIX="s3://${BUCKET}/${PREFIX}/"
readonly COMMIT="${BUILDKITE_COMMIT:-}"
readonly DRY_RUN="${DRY_RUN:-0}"
if [[ "$DRY_RUN" != "0" && "$DRY_RUN" != "1" ]]; then
echo "DRY_RUN must be 0 or 1" >&2
exit 2
fi
if [[ "$DRY_RUN" == "0" && ! "$COMMIT" =~ ^[0-9a-f]{40}$ ]]; then
echo "BUILDKITE_COMMIT must be a full lowercase commit hash" >&2
exit 2
fi
cd "$(dirname "${BASH_SOURCE[0]}")/../../.."
@@ -70,9 +75,9 @@ sed 's/import regex as re/import re/' \
# shellcheck disable=SC2086
$PYTHON "$index_generator" \
--version "$PREFIX" \
--wheel-dir "$PREFIX" \
--wheel-dir "$work_dir/$PREFIX" \
--current-objects "$objects_path" \
--output-dir "$index_output_dir" \
--output-dir "$work_dir" \
--comment "XPU Triton shim"
grep -Fq 'href="triton/"' "$index_output_dir/index.html"
@@ -87,4 +92,7 @@ if [[ "$DRY_RUN" == "1" ]]; then
else
aws s3 cp --recursive "$index_output_dir/" "$S3_PREFIX"
echo "Published XPU Triton shim index to https://wheels.vllm.ai/$PREFIX/"
fi
aws s3 cp "$S3_PREFIX$WHEEL_FILENAME" \
"s3://$BUCKET/$COMMIT/$WHEEL_FILENAME"
echo "Staged XPU Triton shim for https://wheels.vllm.ai/$COMMIT/xpu/"
fi
+30 -3
View File
@@ -129,6 +129,35 @@ CMD ["/bin/bash"]
######################### UCX + NIXL BUILD STAGE #########################
# Build UCX and NIXL in a dedicated stage so compiler/autotools layers are
# never included in the final runtime image (mirrors ROCm's build_rixl stage).
######################### WHEEL BUILD STAGE #########################
# Produces a standalone dist/*.whl artifact (used by the release pipeline's
# "Build wheel - XPU" step), independent of the runtime image below.
FROM vllm-base AS vllm-build
ARG GIT_REPO_CHECK=0
RUN --mount=type=cache,target=/root/.cache/uv \
--mount=type=bind,src=requirements/common.txt,target=/workspace/vllm/requirements/common.txt \
--mount=type=bind,src=requirements/xpu.txt,target=/workspace/vllm/requirements/xpu.txt \
uv pip install grpcio-tools protobuf nanobind && \
uv pip install -r /workspace/vllm/requirements/xpu.txt
# Keep source-dependent layers near the end so frequent code-only changes
# don't invalidate heavy dependency layers.
COPY . .
# Drop the pre-built Rust artifacts into the source tree. setup.py detects
# them and ships them as-is, skipping the local Rust build.
COPY --from=rust-build /workspace/vllm/vllm-rs vllm/vllm-rs
COPY --from=rust-build /workspace/vllm/_rust_*.so vllm/
RUN --mount=type=bind,source=.git,target=.git \
if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
RUN --mount=type=cache,target=/root/.cache/uv \
--mount=type=bind,source=.git,target=.git \
VLLM_TARGET_DEVICE=xpu python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38
FROM vllm-base AS ucx-nixl-build
ARG UCX_VERSION=v1.21.0-rc2
@@ -201,9 +230,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
--mount=type=bind,src=requirements/test/xpu.txt,target=/workspace/vllm/requirements/test/xpu.txt \
uv pip install grpcio-tools protobuf nanobind && \
uv pip install -r /workspace/vllm/requirements/xpu.txt && \
uv pip install --no-build-isolation -r /workspace/vllm/requirements/test/xpu.txt && \
uv pip uninstall triton triton-xpu && \
uv pip install triton-xpu==3.7.2
uv pip install --no-build-isolation -r /workspace/vllm/requirements/test/xpu.txt
# Keep source-dependent layers near the end so frequent code-only changes
# don't invalidate heavy dependency and UCX/NIXL layers.
@@ -20,7 +20,26 @@ There is no extra information on creating a new Python environment for this devi
--8<-- [end:set-up-using-python]
--8<-- [start:pre-built-wheels]
Currently, there are no pre-built XPU wheels.
Pre-built vLLM XPU wheels are published to `wheels.vllm.ai`. Each XPU wheel
index also contains the `triton==3.7.2+xpu` shim described below. PyTorch XPU
packages are served from the PyTorch XPU index, so both index URLs are needed.
#### Install the latest code
To install the wheel built from the latest main branch:
```bash
uv pip install vllm --extra-index-url https://wheels.vllm.ai/nightly/xpu --extra-index-url https://download.pytorch.org/whl/xpu --index-strategy unsafe-best-match
```
#### Install specific revisions
If you want to access the wheels for previous commits (e.g. to bisect the behavior change, performance regression), you can specify the commit hash in the URL:
```bash
export VLLM_COMMIT=730bd35378bf2a5b56b6d3a45be28b3092d26519 # use full commit hash from the main branch
uv pip install vllm --extra-index-url https://wheels.vllm.ai/${VLLM_COMMIT}/xpu --extra-index-url https://download.pytorch.org/whl/xpu --index-strategy unsafe-best-match
```
--8<-- [end:pre-built-wheels]
--8<-- [start:build-wheel-from-source]
@@ -36,25 +55,24 @@ pip install --upgrade pip
pip install -v -r requirements/xpu.txt
```
- Then, install the correct Triton package for Intel XPU.
The default `triton` package (for NVIDIA GPUs) may be installed as a transitive dependency (e.g., via `xgrammar`). For Intel XPU, you must replace it with `triton-xpu`:
```bash
pip uninstall -y triton triton-xpu
pip install triton-xpu==3.7.2 --extra-index-url https://download.pytorch.org/whl/xpu
```
!!! note
- `triton` (without suffix) is for NVIDIA GPUs only. On XPU, using it instead of `triton-xpu` can cause correctness or runtime issues.
- For torch 2.13 (the version used in `requirements/xpu.txt`), the matching package is `triton-xpu==3.7.2`. If you use a different version of torch, check the corresponding `triton-xpu` version in [docker/Dockerfile.xpu](https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile.xpu).
- Finally, build and install vLLM XPU backend:
- Then, install vLLM XPU backend:
```bash
VLLM_TARGET_DEVICE=xpu pip install --no-build-isolation -e . -v
```
!!! note
`requirements/xpu.txt` pins `triton==3.7.2+xpu`, a compatibility shim
hosted on `https://wheels.vllm.ai/xpu` that transparently resolves to
the real Intel XPU implementation (`triton-xpu`). This exists because
some transitive dependencies (e.g. `xgrammar`) unconditionally
require a distribution literally named `triton`, which otherwise
resolves to the NVIDIA-only PyPI `triton` package on XPU and can
cause correctness or runtime issues. No manual uninstall/reinstall of
`triton`/`triton-xpu` is needed; both `pip install` and `uv pip
install --index-strategy unsafe-best-match` resolve the correct
package automatically.
--8<-- [end:build-wheel-from-source]
--8<-- [start:pre-built-images]