Dockerfile

FROM harshmanvar/vllm-cpu-only:v1

# Set writable Hugging Face cache directory
ENV TRANSFORMERS_CACHE=/workspace/hf_cache
ENV HF_HOME=/workspace/hf_cache
RUN mkdir -p /workspace/hf_cache && chmod -R 777 /workspace/hf_cache

RUN pip show vllm

RUN pip list

# Step 10 — Start API server with a model from HF Hub
CMD ["python", "-m", "vllm.entrypoints.openai.api_server", \
     "--model", "unsloth/Llama-3.2-3B-bnb-4bit", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "--trust-remote-code"]