igitman's picture
inference_endpoint (#2)
b1a665d verified
raw
history blame contribute delete
689 Bytes
FROM igitman/nemo-skills-vllm:0.6.0 as base
# Install NeMo-Skills and dependencies
RUN git clone https://github.com/NVIDIA/NeMo-Skills \
&& cd NeMo-Skills \
&& pip install --ignore-installed blinker \
&& pip install -e . \
&& pip install -r requirements/code_execution.txt
# Ensure python is available
RUN ln -s /usr/bin/python3 /usr/bin/python
# Copy our custom files
COPY handler.py server.py /usr/local/endpoint/
# Expose port 80
EXPOSE 80
# Copy and set up entrypoint script
COPY entrypoint.sh /usr/local/endpoint/
RUN chmod +x /usr/local/endpoint/entrypoint.sh
# Set working directory
WORKDIR /usr/local/endpoint
ENTRYPOINT ["/usr/local/endpoint/entrypoint.sh"]