rebase: target the official vLLM 0.31.0 release

Use upstream release images at db9527a46873, preserve custom processor conditioning, and include weekly jobs in full Buildkite release validation.

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
This commit is contained in:
tzhouam committed 2026-10-03 21:11:54 +08:00
1 parent 008fa25263
commit ed876de879
15 files changed
+142 -45

No files matched your search

+8 -10
View File
@@ -1,11 +1,9 @@
# Target an unreleased vLLM main commit. The official commit image carries
# the matching CUDA/PyTorch/FlashInfer stack; do not overlay a nightly wheel
# on the previous release image. Copilot updates this commit pin, which also
# selects the image used by CI. The selected commit must have a published
# nightly image as well as a cu130 wheel.
ARG VLLM_PRECOMPILED_WHEEL_COMMIT=ac9126e58aa7bbab1856ba6593ba4d5003fea516
ARG VLLM_BASE_IMAGE=vllm/vllm-openai
ARG VLLM_BASE_TAG=nightly-${VLLM_PRECOMPILED_WHEEL_COMMIT}
# Use the official release image with its matching CUDA/PyTorch stack.
# The release images are available in upstream ECR before Docker Hub promotion.
ARG VLLM_VERSION=0.31.0
ARG VLLM_PRECOMPILED_WHEEL_COMMIT=db9527a46873454610df6dbedf79a36d6bf1a7f6
ARG VLLM_BASE_IMAGE=public.ecr.aws/q9t5s3a7/vllm-release-repo
ARG VLLM_BASE_TAG=${VLLM_PRECOMPILED_WHEEL_COMMIT}
# We use registry-based layer caching for the dependencies,
# where the tag is determined based on the hash of the contents
@@ -13,7 +11,7 @@ ARG VLLM_BASE_TAG=nightly-${VLLM_PRECOMPILED_WHEEL_COMMIT}
# everything except the actual project dir, since that should not
# change much across branches.
FROM ${VLLM_BASE_IMAGE}:${VLLM_BASE_TAG} AS deps
ARG VLLM_PRECOMPILED_WHEEL_COMMIT
ARG VLLM_VERSION
# Keep APP_DIR outside /workspace so k8s Buildkite workspace mounts cannot
# shadow the editable install target (#5364). Docker WORKDIR creates /opt if
# needed.
@@ -47,7 +45,7 @@ RUN uv pip install --system -e ".[dev]"
# Fail the image build if dependency resolution replaced the selected vLLM.
# The CPU builder has no NVIDIA driver; CUDA imports are checked at the start
# of each GPU test job by .buildkite/cuda/scripts/check_vllm_runtime.py.
RUN python3 -c "import vllm; assert '${VLLM_PRECOMPILED_WHEEL_COMMIT}'[:8] in vllm.__version__, vllm.__version__"
RUN python3 -c "import vllm; assert vllm.__version__ == '${VLLM_VERSION}', vllm.__version__"
# The final stage just copies the working dir on top of the
# dependencies image. This uses --link so that we can skip
+2 -1
View File
@@ -1,4 +1,5 @@
ARG BASE_IMAGE=vllm/vllm-openai:v0.30.0
# Official vLLM v0.31.0 release image, before Docker Hub tag promotion.
ARG BASE_IMAGE=public.ecr.aws/q9t5s3a7/vllm-release-repo:db9527a46873454610df6dbedf79a36d6bf1a7f6
FROM ${BASE_IMAGE}
ARG COMMON_WORKDIR=/app
+4 -3
View File
@@ -1,7 +1,8 @@
# Keep the ROCm runtime aligned with the vLLM release targeted by
# docker/Dockerfile.ci. A stale base image can build successfully while failing
# at runtime when vLLM-Omni imports APIs added by the newer vLLM release.
ARG BASE_IMAGE=vllm/vllm-openai-rocm:v0.30.0
# Official vLLM v0.31.0 release image, before Docker Hub tag promotion.
ARG BASE_IMAGE=public.ecr.aws/q9t5s3a7/vllm-release-repo:db9527a46873454610df6dbedf79a36d6bf1a7f6-rocm
# Keep the dependency-heavy portion independent from the source tree. The
# registry cache is keyed on these inputs, while the final source copy can be
@@ -12,11 +13,11 @@ FROM ${BASE_IMAGE} AS deps
# BASE_IMAGE pins the default prebuilt runtime when USE_NIGHTLY_BUILD=0.
# VLLM_VERSION_OR_COMMIT_HASH is consumed only by the opt-in source rebuild when
# USE_NIGHTLY_BUILD=1. Keep both aligned with docker/Dockerfile.ci's
# VLLM_BASE_TAG. This is enforced by tests/buildkite/test_rocm_dockerfile.py. An
# release version. This is enforced by tests/buildkite/test_rocm_dockerfile.py. An
# intentional custom source pin must update that regression in the same change,
# and should return to the stable release as soon as one is available.
ARG USE_NIGHTLY_BUILD=0
ARG VLLM_VERSION_OR_COMMIT_HASH=v0.30.0
ARG VLLM_VERSION_OR_COMMIT_HASH=v0.31.0
ARG ARG_PYTORCH_ROCM_ARCH
ENV PYTORCH_ROCM_ARCH=${ARG_PYTORCH_ROCM_ARCH:-${PYTORCH_ROCM_ARCH}}