From 9cbc1e27776fc5a6adfc11d14951b12cacebcbf8 Mon Sep 17 00:00:00 2001 From: biondizzle Date: Wed, 15 Apr 2026 07:41:18 +0000 Subject: [PATCH] well need cuda --- Dockerfile | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/Dockerfile b/Dockerfile index fd4a365..ad71e0f 100644 --- a/Dockerfile +++ b/Dockerfile @@ -10,9 +10,12 @@ 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 -# NO_CUDA_EXT=1 skips CUDA extension build (not needed for Redis backend) +# Reinstall torch with CUDA 13 support, then build LMCache with CUDA extensions 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 && \ - NO_CUDA_EXT=1 pip install --no-cache-dir . && \ + pip install --no-cache-dir . && \ rm -rf /tmp/lmcache