biondizzle/vllm-glm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

force for blackwell

Browse Source
This commit is contained in:
biondizzle
2026-04-15 07:56:50 +00:00
parent 9cbc1e2777
commit 28f9f4c172
1 changed files with 4 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
Dockerfile
View File

@@ -10,12 +10,11 @@ COPY utils.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/utils.py
COPY vllm_patches/hf.py /usr/local/lib/python3.12/dist-packages/vllm/renderers/hf.py
# Install LMCache for KV cache offloading / sharing across nodes
# Reinstall torch with CUDA 13 support, then build LMCache with CUDA extensions
# Build with system CUDA 13.0 for Blackwell (B200)
RUN apt-get update && apt-get install -y git && \
# Reinstall torch with CUDA 13.x support to match base image CUDA version
pip install --no-cache-dir --upgrade torch --index-url https://download.pytorch.org/whl/cu131 && \
# Clone and build LMCache from source with CUDA extensions
git clone --depth 1 https://github.com/LMCache/LMCache.git /tmp/lmcache && \
cd /tmp/lmcache && \
pip install --no-cache-dir . && \
CUDA_HOME=/usr/local/cuda \
TORCH_CUDA_ARCH_LIST="10.0" \
pip install --no-cache-dir --no-build-isolation . && \
rm -rf /tmp/lmcache