iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Frontend][RFC] Rust front-end integration (#40848)

Browse Source 
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
This commit is contained in:
Nick Hill
2026-05-21 12:24:48 +08:00
committed by GitHub
co-authored by Bugen Zhao
parent d97ba29fdc
commit f2ace1d57d
35 changed files with 893 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docker/Dockerfile.rocm
+51
View File
@@ -106,14 +106,61 @@ ONBUILD RUN git clone ${VLLM_REPO} \
&& cd vllm \
&& git fetch -v --prune -- origin ${VLLM_BRANCH} \
&& git checkout FETCH_HEAD \
&& git submodule update --init --recursive \
&& if [ ${VLLM_REPO} != "https://github.com/vllm-project/vllm.git" ] ; then \
git remote add upstream "https://github.com/vllm-project/vllm.git" \
&& git fetch upstream ; fi
FROM fetch_vllm_${REMOTE_VLLM} AS fetch_vllm
# -----------------------
# Rust build stage
# Builds the `vllm-rs` frontend in a dedicated stage so the wheel build stages
# don't need the rust toolchain or protoc. Runs in parallel with the main wheel
# build for faster end-to-end builds.
FROM fetch_vllm AS rust-build
ARG COMMON_WORKDIR
# Fail loudly if the rust submodule was not initialized on the host before
# `docker build`. The rust frontend source is brought in via the fetch_vllm
# stage, so an uninitialized submodule would otherwise produce a confusing
# cargo failure.
RUN if [ ! -f ${COMMON_WORKDIR}/vllm/rust/Cargo.toml ]; then \
echo "ERROR: rust/ submodule is not initialized."; \
echo "Run 'git submodule update --init --recursive' on the host before building."; \
exit 1; \
fi
# protoc is used by tonic-build/prost-build.
RUN apt-get update -q -y && apt-get install -q -y --no-install-recommends \
protobuf-compiler libprotobuf-dev \
&& rm -rf /var/lib/apt/lists/*
# Install rustup; the toolchain itself is pinned by rust/rust-toolchain.toml.
RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | \
sh -s -- -y --profile minimal --default-toolchain none
ENV PATH="/root/.cargo/bin:${PATH}"
# Cap cargo parallelism to avoid exhausting the AMD CI host's open-file limit
# (rustc spawns enough concurrent processes to hit RLIMIT_NOFILE otherwise).
ENV CARGO_BUILD_JOBS=4
# Build the release binary. Cache cargo registry/git, and copy the binary out
# so it persists into the image layer for later COPY --from=rust-build.
RUN --mount=type=cache,target=/root/.cargo/registry \
--mount=type=cache,target=/root/.cargo/git \
cd ${COMMON_WORKDIR}/vllm/rust \
&& cargo build --release --bin vllm-rs --features native-tls-vendored \
&& cp target/release/vllm-rs /tmp/vllm-rs
# -----------------------
# vLLM build stages
FROM fetch_vllm AS build_vllm
ARG COMMON_WORKDIR
# Drop the pre-built rust frontend binary into the source tree. setup.py
# detects it and ships it as-is, skipping the local cargo build.
COPY --from=rust-build /tmp/vllm-rs ${COMMON_WORKDIR}/vllm/vllm/vllm-rs
# Build vLLM (setup.py auto-detects sccache in PATH)
RUN cd vllm \
&& python3 -m pip install -r requirements/rocm.txt \
@@ -293,6 +340,10 @@ FROM fetch_vllm AS build_vllm_wheel_release
ARG COMMON_WORKDIR
# Drop the pre-built rust frontend binary into the source tree. setup.py
# detects it and ships it as-is, skipping the local cargo build.
COPY --from=rust-build /tmp/vllm-rs ${COMMON_WORKDIR}/vllm/vllm/vllm-rs
# Create /install directory for custom wheels
RUN mkdir -p /install