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

Applied new fastsafetensors fix to mxfp4 build; disabled wheel builds by default

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-02-09 23:47:06 -08:00
parent 74876dd442
commit ace16f3a8f
7 changed files with 71 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Dockerfile.mxfp4
View File

@@ -211,9 +211,9 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install -r requirements/build.txt
# Apply Patches
# TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36
#COPY fastsafetensors.patch .
#RUN patch -p1 < fastsafetensors.patch
# TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/vllm-project/vllm/issues/34180
COPY fastsafetensors_mxfp4.patch .
RUN patch -p1 < fastsafetensors_mxfp4.patch
# Final Compilation
# We mount the ccache directory here. Ideally, map this to a host volume for persistence