[fix] add kv routing to vllm dockerfile

container/Dockerfile.vllm

@@ -78,14 +78,26 @@ COPY runtime /workspace/runtime
 RUN cd runtime/rust && \
     cargo build --release --locked && cargo doc --no-deps
 
+# Generate C bindings for kv cache routing in vLLM
+COPY llm /workspace/llm
+RUN cd llm/rust/ && \
+cargo build --release --locked && cargo doc --no-deps
+
 # Build triton_distributed_rs wheel
-RUN cd runtime/rust/python-wheel && \
+COPY python-wheel /workspace/python-wheel
+RUN cd python-wheel && \
     uv build && \
     uv pip install dist/triton_distributed_rs*cp312*.whl
 
+# Package the bindings
+RUN mkdir -p /opt/triton/llm_binding/wheels && mkdir /opt/triton/llm_binding/lib
+RUN cp python-wheel/dist/triton_distributed_rs*cp312*.whl /opt/triton/llm_binding/wheels/.
+RUN cp llm/rust/target/release/libtriton_llm_capi.so /opt/triton/llm_binding/lib/.
+RUN cp -r llm/rust/libtriton-llm/include /opt/triton/llm_binding/.
+
 # Install patched vllm
 ARG VLLM_REF="v0.7.2"
-ARG VLLM_PATCH="vllm_${VLLM_REF}.patch"
+ARG VLLM_PATCH="vllm_${VLLM_REF}-triton-kv-disagg-patch.patch"
 RUN --mount=type=bind,source=./container/deps/,target=/tmp/deps \
     bash /tmp/deps/vllm/install.sh --patch /tmp/deps/vllm/${VLLM_PATCH} --ref ${VLLM_REF} --install-cmd "uv pip install --editable" --use-precompiled --installation-dir /opt/vllm
 
@@ -106,6 +118,8 @@ COPY . /workspace
 # Environment setup
 ENV PYTHONPATH="${PYTHONPATH}:/workspace/examples/python:/opt/tritonserver/python/openai/openai_frontend"
 ENV RAPIDS_LIBUCX_PREFER_SYSTEM_LIBRARY=true
+# Tell vllm to use the Triton LLM C API for KV Cache Routing
+ENV VLLM_KV_CAPI_PATH="/opt/triton/llm_binding/lib/libtriton_llm_capi.so"
 
 CMD []
 ENTRYPOINT ["/opt/nvidia/nvidia_entrypoint.sh"]