FROM harbor.4pd.io/modelhubxc/enginex-mthreads/vllm-musa-qy2-py310:v0.8.4-release

COPY fix_tokenizer.py /opt/fix_tokenizer.py
COPY vllm_wrapper.sh /opt/vllm_wrapper.sh

RUN chmod +x /opt/vllm_wrapper.sh && \
    mv /usr/local/bin/vllm /usr/local/bin/vllm_real && \
    ln -s /opt/vllm_wrapper.sh /usr/local/bin/vllm