turbopilot/Dockerfile.cuda

29 lines
586 B
Docker
Raw Normal View History

2023-05-08 10:38:25 +00:00
FROM nvidia/cuda:11.7.1-devel-ubuntu22.04 AS build
2023-04-10 07:51:48 +00:00
2023-05-08 10:38:25 +00:00
RUN apt-get update && apt-get install -y build-essential cmake libboost-dev
2023-04-10 07:51:48 +00:00
ADD ./ggml /build/ggml
RUN mkdir /build/ggml/build
WORKDIR /build/ggml/build
2023-05-08 10:38:25 +00:00
RUN cmake -DGGML_CUBLAS=ON -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc ..
2023-04-10 07:51:48 +00:00
RUN make codegen-serve
2023-05-08 10:38:25 +00:00
FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu22.04 AS runtime
2023-04-10 07:51:48 +00:00
WORKDIR /app
COPY --from=build /build/ggml/build/bin/codegen-serve /app/codegen-serve
ENV THREADS=4
2023-04-10 08:18:04 +00:00
ENV MODEL="/models/codegen-2B-multi-ggml-4bit-quant.bin"
2023-04-10 07:51:48 +00:00
COPY ./run.sh /app/
EXPOSE 18080
CMD /app/run.sh