patch vLLM

Author: wsmlby

server/Dockerfile.cuda

@@ -18,8 +18,12 @@ RUN pip install -r requirements.txt
 # RUN pip install .
 ENV ACCELERATOR=CUDA
 # Copy our application source code
-COPY vllm_patched /usr/local/lib/python3.12/dist-packages/vllm
 COPY ./homl_server ./homl_server
+COPY ./vllm_patches ./patches
+
+RUN cd /usr/local/lib/python3.12/dist-packages/vllm && patch -p1 < /app/patches/api_server.patch
+RUN cd /usr/local/lib/python3.12/dist-packages/vllm && patch -p1 < /app/patches/registry.patch
+
 WORKDIR /app/homl_server
 # The base image exposes port 8000, so we don't need to do it again.
 # EXPOSE 8000

server/Dockerfile.cuda-gptoss

@@ -18,6 +18,11 @@ RUN pip install -r requirements.txt
 ENV ACCELERATOR=CUDA
 # Copy our application source code
 COPY ./homl_server ./homl_server
+COPY ./vllm_patches ./patches
+
+RUN cd /usr/local/lib/python3.12/dist-packages/vllm && patch -p1 < /app/patches/api_server.patch
+RUN cd /usr/local/lib/python3.12/dist-packages/vllm && patch -p1 < /app/patches/registry.patch
+
 WORKDIR /app/homl_server
 # The base image exposes port 8000, so we don't need to do it again.
 # EXPOSE 8000

server/create_patch.sh

@@ -0,0 +1,2 @@
+diff -u vllm_origin/entrypoints/openai/api_server.py vllm_patched/entrypoints/openai/api_server.py > vllm_patches/api_server.patch
+diff -u vllm_origin/model_executor/models/registry.py vllm_patched/model_executor/models/registry.py > vllm_patches/registry.patch