software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Reverted back to release flashinfer

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2025-12-20 23:01:49 -08:00
parent d5d85aaac7
commit 9f35dbdd2d
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Dockerfile.wheels
View File

@@ -63,9 +63,9 @@ RUN patch -p1 -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fast
# Install flashinfer helper packages
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install --system --break-system-packages flashinfer-python -U --no-deps --pre --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages flashinfer-cubin --pre --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages flashinfer-jit-cache --pre --index-url https://flashinfer.ai/whl/cu130
uv pip install --system --break-system-packages flashinfer-python -U --no-deps --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages flashinfer-cubin --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu130
# Setup Env for Runtime
ENV TORCH_CUDA_ARCH_LIST=12.1a