[Build] Fix CUDA arch build coverage gaps (#45277)

Signed-off-by: Shengqi Chen <harry-chen@outlook.com>
Co-authored-by: Xin Li <xinli-sw@users.noreply.github.com>
Co-authored-by: ShawRong <ShawRong@users.noreply.github.com>
Co-authored-by: Change72 <Change72@users.noreply.github.com>
This commit is contained in:
Shengqi Chen
2026-06-13 22:09:20 -07:00
committed by GitHub
co-authored by Xin Li ShawRong Change72
parent 54bbf51668
commit 78e7293bb1
14 changed files with 240 additions and 132 deletions
+6 -2
View File
@@ -261,7 +261,10 @@ RUN --mount=type=cache,target=/opt/uv/cache \
# Explicitly set the list to avoid issues with torch 2.2
# See https://github.com/pytorch/pytorch/pull/123243
# From versions.json: .torch.cuda_arch_list
ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX'
# Do not add +PTX here: vLLM filters torch's top-level PTX flag when it
# converts global gencode flags into per-kernel arch lists. If a specific
# kernel needs PTX, add +PTX to that kernel's CMake arch list instead.
ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
#################### BUILD BASE IMAGE ####################
@@ -1010,7 +1013,8 @@ ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
ENV UV_HTTP_TIMEOUT=500
# install kv_connectors if requested
ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX'
# Do not add +PTX here; see the main TORCH_CUDA_ARCH_LIST comment above.
ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
RUN --mount=type=cache,target=/opt/uv/cache \
--mount=type=bind,source=requirements/kv_connectors.txt,target=/tmp/kv_connectors.txt,ro \
+1 -1
View File
@@ -35,7 +35,7 @@
"default": "false"
},
"TORCH_CUDA_ARCH_LIST": {
"default": "7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX"
"default": "7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0"
},
"MAX_JOBS": {
"default": "2"