diff --git a/Dockerfile b/Dockerfile
index 8d9b8ea..4f6f11c 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -10,4 +10,8 @@ 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
-RUN pip install --no-cache-dir lmcache
+# Build from source to match CUDA 13 (pip wheel is CUDA 12)
+RUN git clone --depth 1 https://github.com/LMCache/LMCache.git /tmp/lmcache && \
+    cd /tmp/lmcache && \
+    pip install --no-cache-dir --no-build-isolation -e . && \
+    rm -rf /tmp/lmcache