34 lines
1001 B
Docker
34 lines
1001 B
Docker
FROM docker.1ms.run/kyuz0/vllm-therock-gfx1201:latest
|
|
|
|
# Set environment variables
|
|
ENV USE_DEFAULT_MODEL=true
|
|
ENV LOCAL_MODEL_DIR=/opt/model
|
|
|
|
# Create necessary directories
|
|
RUN mkdir -p /opt/script /opt/model /config
|
|
|
|
# Copy scripts to /opt/script
|
|
COPY scripts/add_qwen3_5_moe_support.py /opt/script/add_qwen3_5_moe_support.py
|
|
COPY scripts/start_vllm.py /opt/script/start-vllm
|
|
COPY benchmarks/run_vllm_bench.py /opt/script/run_vllm_bench.py
|
|
|
|
# Update vLLM to nightly version for Qwen3.5 support
|
|
RUN pip install --upgrade vllm --extra-index-url https://wheels.vllm.ai/nightly
|
|
|
|
# Add qwen3_5_moe support to Transformers
|
|
RUN python3 /opt/script/add_qwen3_5_moe_support.py
|
|
|
|
# Note: config.json should be mounted at runtime via docker-compose.yml
|
|
|
|
# Make scripts executable
|
|
RUN chmod +x /opt/script/start-vllm
|
|
|
|
# Replace the container's start-vllm with our improved version
|
|
RUN cp /opt/script/start-vllm /usr/local/bin/start-vllm
|
|
|
|
# Set working directory
|
|
WORKDIR /opt
|
|
|
|
# Default command
|
|
CMD ["start-vllm"]
|