Dockerfile

FROM nvidia/cuda:11.8.0-devel-ubuntu22.04

# Install necessary packages
RUN apt update && apt install git build-essential libopenblas-dev wget python3-pip -y
RUN apt update && apt install -y wget

# Set up the working directory
ARG MODEL_PART_A
ARG MODEL_PART_B
ARG MODEL_NAME
ARG ADDITIONAL
RUN git clone https://github.com/theroyallab/tabbyAPI
WORKDIR /tabbyAPI
RUN pip install -r requirements.txt -q
RUN pip install huggingface-hub -q
RUN echo "network:" >> config.yml && \
    echo "  host: 127.0.0.1" >> config.yml && \
    echo "  port: 5000" >> config.yml && \
    echo "  disable_auth: False" >> config.yml && \
    echo "" >> config.yml && \
    echo "logging:" >> config.yml && \
    echo "  prompt: False" >> config.yml && \
    echo "  generation_params: False" >> config.yml && \
    echo "" >> config.yml && \
    echo "sampling:" >> config.yml && \
    echo "  override_preset: null" >> config.yml && \
    echo "" >> config.yml && \
    echo "developer:" >> config.yml && \
    echo "  unsafe_launch: False" >> config.yml && \
    echo "" >> config.yml && \
    echo "model:" >> config.yml && \
    echo "  model_dir: models" >> config.yml && \
    echo "  model_name: null" >> config.yml && \
    echo "  use_dummy_models: False" >> config.yml
CMD ["python3", "main.py"]