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

10
Dockerfile.wheels
View File

@@ -40,7 +40,7 @@ RUN mkdir -p tiktoken_encodings && \
wget -O tiktoken_encodings/o200k_base.tiktoken "https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken" && \
wget -O tiktoken_encodings/cl100k_base.tiktoken "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken"
# Cache TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36
# Cache TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/vllm-project/vllm/issues/34180
# COPY fastsafetensors.patch .
# Install fastsafetensors
@@ -66,9 +66,13 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
--extra-index-url https://wheels.vllm.ai/nightly/cu130; \
fi
# Apply TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/foundation-model-stack/fastsafetensors/issues/36
# Apply TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/vllm-project/vllm/issues/34180
# Apply in site-packages
# RUN patch -p1 -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fastsafetensors.patch
# RUN if patch -p1 --dry-run --reverse -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fastsafetensors.patch &>/dev/null; then \
# echo "PR #34180 is already applied"; \
# else \
# patch -p1 -d /usr/local/lib/python3.12/dist-packages < ${VLLM_BASE_DIR}/fastsafetensors.patch; \
# fi
ARG FLASHINFER_PRE=""