From 28f9f4c17249891d1d0a488526daac968face31d Mon Sep 17 00:00:00 2001 From: biondizzle Date: Wed, 15 Apr 2026 07:56:50 +0000 Subject: [PATCH] force for blackwell --- Dockerfile | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/Dockerfile b/Dockerfile index ad71e0f..a4acb87 100644 --- a/Dockerfile +++ b/Dockerfile @@ -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