docker/Dockerfile.jetson

#Standalone Dockerfile for text-generation-webui on NVIDIA Jetson Embedded devices

FROM nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3 as builder
ENV TORCH_CUDA_ARCH_LIST Turing
RUN apt-get update && \
    apt-get install -y python3 python3-pip git build-essential python3-dev

RUN pip3 install --upgrade pip setuptools
RUN git clone https://github.com/g588928812/bitsandbytes_jetsonX.git /build
WORKDIR /build
RUN CUDA_VERSION=118 make cuda11x
RUN mkdir /wheels
RUN python3 setup.py bdist_wheel -d /wheels
RUN rm -rf /build
RUN git clone https://github.com/oobabooga/GPTQ-for-LLaMa /build
WORKDIR /build
RUN pip3 install -r requirements.txt
RUN python3 setup_cuda.py bdist_wheel -d /wheels

FROM nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3
COPY --from=builder /wheels /wheels
COPY --from=builder /build /build
RUN apt-get update && \
    apt-get install --no-install-recommends -y git python3-dev python3 python3-pip make g++ && \
    rm -rf /var/lib/apt/lists/*
RUN pip3 install /wheels/*.whl
RUN rm -rf /wheels
WORKDIR /build
RUN pip3 install -r requirements.txt
RUN git clone https://github.com/oobabooga/text-generation-webui /app

WORKDIR /app
#ENV WEBUI_VERSION="2908a515877ffde2b1684b2353f6d72e6cb4d31b"
#RUN git reset --hard ${WEBUI_VERSION}
RUN pip3 install --upgrade pip setuptools
RUN pip3 install protobuf>=3.3.0
RUN pip3 install -r requirements.txt
#Force to use bitsandbytes_jetsonX
RUN pip3 uninstall -y bitsandbytes
RUN mkdir /app/repositories
RUN mv /build /app/repositories/GPTQ-for-LLaMa

#Remove Python 3.10 specific macros
RUN sed -i 's/@functools.cache/@functools.lru_cache(maxsize=None)/g' /app/modules/chat.py
RUN sed -i 's/@functools.cache/@functools.lru_cache(maxsize=None)/g' /app/modules/loaders.py
RUN sed -i 's/@functools.cache/@functools.lru_cache(maxsize=None)/g' /app/modules/presets.py

EXPOSE 7860

ENV CLI_ARGS="--listen"
CMD python3 server.py ${CLI_ARGS}