Skip to content

Commit 2846d06

Browse files
Add HPU image build with vllm-gaudi compatibility
Signed-off-by: jakub-sochacki <jakub.sochacki@intel.com>
1 parent 5fbb4c3 commit 2846d06

File tree

1 file changed

+27
-5
lines changed

1 file changed

+27
-5
lines changed

buildkite/test-template-ci.j2

Lines changed: 27 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,8 @@
99
{% set docker_image_torch_nightly = "public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT-torch-nightly" %}
1010
{% set docker_image_cu118 = "public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT-cu118" %}
1111
{% set docker_image_cpu = "public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT-cpu" %}
12-
{% set docker_image_hpu = "public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT-hpu" %}
12+
{# Note: docker_image_hpu is NOT defined here because HPU uses a different commit tag (from vllm-gaudi compatibility) #}
13+
{# HPU image tag is determined dynamically in the build step based on VLLM_STABLE_COMMIT #}
1314
{% endif %}
1415
{% set docker_image_amd = "rocm/vllm-ci:$BUILDKITE_COMMIT" %}
1516
{% set default_working_dir = "/vllm-workspace/tests" %}
@@ -412,6 +413,7 @@ steps:
412413
- exit_status: -10 # Agent was lost
413414
limit: 2
414415

416+
{% if branch == "main" %}
415417
- label: ":docker: build image HPU"
416418
key: image-build-hpu
417419
depends_on: ~
@@ -421,14 +423,33 @@ steps:
421423
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
422424
- |
423425
#!/bin/bash
424-
if [[ -z $(docker manifest inspect {{ docker_image_hpu }}) ]]; then
426+
# Fetch the compatible vLLM commit for vllm-gaudi
427+
VLLM_STABLE_COMMIT=$(curl -s https://raw.githubusercontent.com/vllm-project/vllm-gaudi/main/last-good-commit-for-vllm-gaudi/VLLM_STABLE_COMMIT | tr -d '\n')
428+
echo "Compatible vLLM commit for vllm-gaudi: $VLLM_STABLE_COMMIT"
429+
430+
# HPU images always use postmerge registry (main branch only)
431+
REGISTRY="public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo"
432+
433+
# HPU images use the stable commit tag, not BUILDKITE_COMMIT
434+
HPU_IMAGE_TAG="${REGISTRY}:${VLLM_STABLE_COMMIT}-hpu"
435+
436+
if [[ -z $(docker manifest inspect "$HPU_IMAGE_TAG") ]]; then
425437
echo "Image not found, proceeding with build..."
426438
else
427-
echo "Image found"
439+
echo "Image $HPU_IMAGE_TAG already exists"
428440
exit 0
429441
fi
430-
- "docker build --file docker/Dockerfile.hpu --build-arg max_jobs=16 --build-arg buildkite_commit=$BUILDKITE_COMMIT --tag {{ docker_image_hpu }} --progress plain ."
431-
- "docker push {{ docker_image_hpu }}"
442+
443+
# Build with the stable commit
444+
docker build \
445+
--file docker/Dockerfile.hpu \
446+
--build-arg max_jobs=16 \
447+
--build-arg VLLM_COMMIT=$VLLM_STABLE_COMMIT \
448+
--build-arg VLLM_GAUDI_COMMIT=main \
449+
--tag "$HPU_IMAGE_TAG" \
450+
--progress plain .
451+
452+
docker push "$HPU_IMAGE_TAG"
432453
env:
433454
DOCKER_BUILDKIT: "1"
434455
retry:
@@ -437,6 +458,7 @@ steps:
437458
limit: 2
438459
- exit_status: -10 # Agent was lost
439460
limit: 2
461+
{% endif %}
440462

441463
{% for step in steps %}
442464
{% if step.fast_check_only != true %}

0 commit comments

Comments
 (0)