[CI/Build][Docker] Bump nvidia-cutlass-dsl to 4.6.0 and drop packaging workarounds (#47442)

Signed-off-by: Artem Perevedentsev <aperevedents@nvidia.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Artem Perevedentsev
2026-07-16 13:51:41 +00:00
committed by GitHub
co-authored by Matthew Bonanni
parent b8168e33e0
commit 7cd1d57b74
7 changed files with 8 additions and 120 deletions
-35
View File
@@ -248,10 +248,6 @@ COPY requirements/common.txt requirements/common.txt
COPY requirements/cuda.txt requirements/cuda.txt
COPY use_existing_torch.py use_existing_torch.py
COPY pyproject.toml pyproject.toml
# nvidia-cutlass-dsl[cu13] installs -libs-base and -libs-cu13 wheels that
# share paths with different content. uv can extract them in either order,
# leaving base files that break CUDA 13 CuTe DSL JIT.
# TODO(mmangkad): Remove this after NVIDIA/cutlass#3259 is fixed.
RUN --mount=type=cache,target=/opt/uv/cache \
if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "12" ]; then \
sed -i 's/^nvidia-cutlass-dsl\[cu13\]/nvidia-cutlass-dsl/' requirements/cuda.txt; \
@@ -268,13 +264,6 @@ RUN --mount=type=cache,target=/opt/uv/cache \
else \
uv pip install --python /opt/venv/bin/python3 -r requirements/cuda.txt \
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); \
fi \
&& if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "13" ]; then \
CUTLASS_DSL_VERSION=$(uv pip show --python /opt/venv/bin/python3 nvidia-cutlass-dsl 2>/dev/null | awk '/^Version:/{print $2}') && \
if [ -n "$CUTLASS_DSL_VERSION" ]; then \
uv pip install --python /opt/venv/bin/python3 --force-reinstall --no-deps \
"nvidia-cutlass-dsl-libs-cu13==${CUTLASS_DSL_VERSION}"; \
fi; \
fi
# Track PyTorch lib versions used during build and match in downstream instances.
@@ -792,10 +781,6 @@ ENV VLLM_ENABLE_CUDA_COMPATIBILITY=0
ARG PYTORCH_CUDA_INDEX_BASE_URL
COPY requirements/common.txt /tmp/common.txt
COPY requirements/cuda.txt /tmp/requirements-cuda.txt
# nvidia-cutlass-dsl[cu13] installs -libs-base and -libs-cu13 wheels that
# share paths with different content. uv can extract them in either order,
# leaving base files that break CUDA 13 CuTe DSL JIT.
# TODO(mmangkad): Remove this after NVIDIA/cutlass#3259 is fixed.
RUN --mount=type=cache,target=/opt/uv/cache \
if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "12" ]; then \
sed -i 's/^nvidia-cutlass-dsl\[cu13\]/nvidia-cutlass-dsl/' /tmp/requirements-cuda.txt; \
@@ -803,13 +788,6 @@ RUN --mount=type=cache,target=/opt/uv/cache \
fi && \
uv pip install --system -r /tmp/requirements-cuda.txt \
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') && \
if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "13" ]; then \
CUTLASS_DSL_VERSION=$(uv pip show --system nvidia-cutlass-dsl 2>/dev/null | awk '/^Version:/{print $2}') && \
if [ -n "$CUTLASS_DSL_VERSION" ]; then \
uv pip install --system --force-reinstall --no-deps \
"nvidia-cutlass-dsl-libs-cu13==${CUTLASS_DSL_VERSION}"; \
fi; \
fi && \
rm /tmp/requirements-cuda.txt /tmp/common.txt
# Install FlashInfer JIT cache (requires CUDA-version-specific index URL)
@@ -908,19 +886,6 @@ RUN --mount=type=bind,from=build,src=/tmp/ep_kernels_workspace/dist,target=/vllm
uv pip install --system ep_kernels/dist/*.whl --verbose \
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
# nvidia-cutlass-dsl[cu13] installs -libs-base and -libs-cu13 wheels that
# share paths with different content. Force -libs-cu13 last after runtime
# dependency installs so uv cannot leave base files behind.
# TODO(mmangkad): Remove this after NVIDIA/cutlass#3259 is fixed.
RUN --mount=type=cache,target=/opt/uv/cache \
if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "13" ]; then \
CUTLASS_DSL_VERSION=$(uv pip show --system nvidia-cutlass-dsl 2>/dev/null | awk '/^Version:/{print $2}') && \
if [ -n "$CUTLASS_DSL_VERSION" ]; then \
uv pip install --system --force-reinstall --no-deps \
"nvidia-cutlass-dsl-libs-cu13==${CUTLASS_DSL_VERSION}"; \
fi; \
fi
# CUDA image changed from /usr/local/nvidia to /usr/local/cuda in 12.8 but will
# return to /usr/local/nvidia in 13.0 to allow container providers to mount drivers
# consistently from the host (see https://github.com/vllm-project/vllm/issues/18859).