[Rebase] Align vLLM-Omni with vLLM v0.31.0 (#8459)

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
This commit is contained in:
Zhou Taichang authored and GitHub committed 2026-10-05 14:11:01 +08:00
1 parent 1f47fa2f29
commit 0b0d2d7009
141 files changed
+2885 -721

No files matched your search

+10 -7
View File
@@ -1,10 +1,7 @@
# Target vLLM version: v0.30.0 (tag @ ced6857afa0e).
# Use the matching release image and its coherent CUDA/PyTorch/FlashInfer
# runtime. Do not reinstall a commit wheel or repair dependencies on top of
# an older base image at a release boundary. For a future unreleased target,
# restore an explicit wheel pin and validate its complete runtime stack.
# Use the official release image with its matching CUDA/PyTorch stack.
ARG VLLM_VERSION=0.31.0
ARG VLLM_BASE_IMAGE=vllm/vllm-openai
ARG VLLM_BASE_TAG=v0.30.0
ARG VLLM_BASE_TAG=v${VLLM_VERSION}
# We use registry-based layer caching for the dependencies,
# where the tag is determined based on the hash of the contents
@@ -12,6 +9,7 @@ ARG VLLM_BASE_TAG=v0.30.0
# 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_VERSION
# Keep APP_DIR outside /workspace so k8s Buildkite workspace mounts cannot
# shadow the editable install target (#5364). Docker WORKDIR creates /opt if
# needed.
@@ -36,12 +34,17 @@ RUN apt-get update && \
# for CI
COPY pyproject.toml setup.py ./
COPY requirements/ requirements/
ENV SETUPTOOLS_SCM_PRETEND_VERSION=0.0.0
# Scope the override to Omni; dependency source builds need their own versions.
ENV VLLM_OMNI_VERSION_OVERRIDE=0.0.0
# Editable install: deps + project metadata are cached in this layer, while
# the package itself is linked to APP_DIR (populated later by COPY --link).
# Do NOT use a non-editable install here — that copies an empty stub into
# site-packages and breaks ``import vllm_omni.entrypoints``.
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.__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