install vllm from source

Dockerfile

@@ -2,6 +2,12 @@ FROM python:3.10-slim
 
 WORKDIR /app
 
+# Environment variables for CPU-only operation
+ENV CUDA_VISIBLE_DEVICES=-1 \
+    TF_CPP_MIN_LOG_LEVEL=3 \
+    SDL_AUDIODRIVER=disk \
+    VLLM_TARGET_DEVICE=cpu
+
 # Install system dependencies (probably more than needed)
 RUN apt-get update && apt-get install -y \
     build-essential \
@@ -56,6 +62,15 @@ RUN pip install \
   transformers \
   --extra-index-url https://download.pytorch.org/whl/cpu
 
+# Install vLLM CPU version from source
+RUN git clone https://github.com/vllm-project/vllm.git \
+    && cd vllm \
+    && pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu \
+    && python setup.py install \
+    && pip install --no-cache-dir -e . \
+    && cd .. \
+    && rm -rf vllm
+
 # Fix networkx compatibility
 RUN pip install --force-reinstall networkx==3.2.1
 
@@ -69,11 +84,5 @@ RUN python -m unidic download
 # Copy application files
 COPY . .
 
-# Environment variables for CPU-only operation
-ENV CUDA_VISIBLE_DEVICES=-1 \
-    TF_CPP_MIN_LOG_LEVEL=3 \
-    SDL_AUDIODRIVER=disk \
-    VLLM_TARGET_DEVICE=cpu
-
 EXPOSE 7860
 CMD ["python", "app.py"]