biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CPU][ARM] Add ARM BF16 cross-compilation support and improve documen… (#33079)

Browse Source 
Signed-off-by: Maryam Tahhan <mtahhan@redhat.com>
Co-authored-by: Li, Jiang <jiang1.li@intel.com>
This commit is contained in:
Maryam Tahhan
2026-02-15 14:33:08 +00:00
committed by GitHub
parent 71cd89264f
commit f07a128413
3 changed files with 81 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
docker/Dockerfile.cpu
View File

@@ -20,6 +20,7 @@
# VLLM_CPU_AVX512BF16=false (default)|true (for cross-compilation)
# VLLM_CPU_AVX512VNNI=false (default)|true (for cross-compilation)
# VLLM_CPU_AMXBF16=false (default)|true (for cross-compilation)
# VLLM_CPU_ARM_BF16=false (default)|true (for cross-compilation)
#
######################### COMMON BASE IMAGE #########################
@@ -108,9 +109,22 @@ ENV VLLM_CPU_AVX512VNNI=${VLLM_CPU_AVX512VNNI}
# Support for building with AMXBF16 ISA: docker build --build-arg VLLM_CPU_AMXBF16="true" ...
ARG VLLM_CPU_AMXBF16=1
ENV VLLM_CPU_AMXBF16=${VLLM_CPU_AMXBF16}
# Support for cross-compilation with ARM BF16 ISA: docker build --build-arg VLLM_CPU_ARM_BF16="true" ...
ARG VLLM_CPU_ARM_BF16=0
ENV VLLM_CPU_ARM_BF16=${VLLM_CPU_ARM_BF16}
WORKDIR /vllm-workspace
# Validate build arguments - prevent mixing incompatible ISA flags
RUN if [ "$TARGETARCH" = "arm64" ] && { [ "$VLLM_CPU_AVX2" != "0" ] || [ "$VLLM_CPU_AVX512" != "0" ] || [ "$VLLM_CPU_AVX512BF16" != "0" ] || [ "$VLLM_CPU_AVX512VNNI" != "0" ]; }; then \
echo "ERROR: Cannot use x86-specific ISA flags (AVX2, AVX512, etc.) when building for ARM64 (--platform=linux/arm64)"; \
exit 1; \
fi && \
if [ "$TARGETARCH" = "amd64" ] && [ "$VLLM_CPU_ARM_BF16" != "0" ]; then \
echo "ERROR: Cannot use ARM-specific ISA flags (ARM_BF16) when building for x86_64 (--platform=linux/amd64)"; \
exit 1; \
fi
# Copy build requirements
COPY requirements/cpu-build.txt requirements/build.txt
@@ -224,6 +238,7 @@ ARG VLLM_CPU_AVX512
ARG VLLM_CPU_AVX512BF16
ARG VLLM_CPU_AVX512VNNI
ARG VLLM_CPU_AMXBF16
ARG VLLM_CPU_ARM_BF16
ARG PYTHON_VERSION
LABEL ai.vllm.build.target-arch="${TARGETARCH}"
@@ -233,6 +248,7 @@ LABEL ai.vllm.build.cpu-avx512="${VLLM_CPU_AVX512:-false}"
LABEL ai.vllm.build.cpu-avx512bf16="${VLLM_CPU_AVX512BF16:-false}"
LABEL ai.vllm.build.cpu-avx512vnni="${VLLM_CPU_AVX512VNNI:-false}"
LABEL ai.vllm.build.cpu-amxbf16="${VLLM_CPU_AMXBF16:-false}"
LABEL ai.vllm.build.cpu-arm-bf16="${VLLM_CPU_ARM_BF16:-false}"
LABEL ai.vllm.build.python-version="${PYTHON_VERSION:-3.12}"
ENTRYPOINT ["vllm", "serve"]