diff --git a/.devops/full-cuda.Dockerfile b/.devops/full-cuda.Dockerfile index 61f671465..b8a354246 100644 --- a/.devops/full-cuda.Dockerfile +++ b/.devops/full-cuda.Dockerfile @@ -1,18 +1,16 @@ ARG UBUNTU_VERSION=22.04 - # This needs to generally match the container host's environment. -ARG CUDA_VERSION=11.7.1 - +ARG CUDA_VERSION=12.6.0 # Target the CUDA build image ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION} FROM ${BASE_CUDA_DEV_CONTAINER} AS build -# Unless otherwise specified, we make a fat build. -ARG CUDA_DOCKER_ARCH=all +# CUDA architecture to build for (defaults to all supported archs) +ARG CUDA_DOCKER_ARCH=default RUN apt-get update && \ - apt-get install -y build-essential python3 python3-pip git libcurl4-openssl-dev libgomp1 + apt-get install -y build-essential cmake python3 python3-pip git libcurl4-openssl-dev libgomp1 COPY requirements.txt requirements.txt COPY requirements requirements @@ -24,13 +22,12 @@ WORKDIR /app COPY . . -# Set nvcc architecture -ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH} -# Enable CUDA -ENV GGML_CUDA=1 -# Enable cURL -ENV LLAMA_CURL=1 - -RUN make -j$(nproc) +# Use the default CUDA archs if not specified +RUN if [ "${CUDA_DOCKER_ARCH}" != "default" ]; then \ + export CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CUDA_DOCKER_ARCH}"; \ + fi && \ + cmake -B build -DGGML_CUDA=ON -DLLAMA_CURL=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \ + cmake --build build --config Release --target llama-cli -j$(nproc) && \ + cp build/bin/* . ENTRYPOINT ["/app/.devops/tools.sh"] diff --git a/.devops/llama-cli-cuda.Dockerfile b/.devops/llama-cli-cuda.Dockerfile index 8eda63a89..b75163b94 100644 --- a/.devops/llama-cli-cuda.Dockerfile +++ b/.devops/llama-cli-cuda.Dockerfile @@ -1,6 +1,6 @@ ARG UBUNTU_VERSION=22.04 # This needs to generally match the container host's environment. -ARG CUDA_VERSION=11.7.1 +ARG CUDA_VERSION=12.6.0 # Target the CUDA build image ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION} # Target the CUDA runtime image @@ -8,28 +8,30 @@ ARG BASE_CUDA_RUN_CONTAINER=nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_V FROM ${BASE_CUDA_DEV_CONTAINER} AS build -# Unless otherwise specified, we make a fat build. -ARG CUDA_DOCKER_ARCH=all +# CUDA architecture to build for (defaults to all supported archs) +ARG CUDA_DOCKER_ARCH=default RUN apt-get update && \ - apt-get install -y build-essential git + apt-get install -y build-essential git cmake WORKDIR /app COPY . . -# Set nvcc architecture -ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH} -# Enable CUDA -ENV GGML_CUDA=1 - -RUN make -j$(nproc) llama-cli +# Use the default CUDA archs if not specified +RUN if [ "${CUDA_DOCKER_ARCH}" != "default" ]; then \ + export CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CUDA_DOCKER_ARCH}"; \ + fi && \ + cmake -B build -DGGML_CUDA=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \ + cmake --build build --config Release --target llama-cli -j$(nproc) FROM ${BASE_CUDA_RUN_CONTAINER} AS runtime RUN apt-get update && \ apt-get install -y libgomp1 -COPY --from=build /app/llama-cli /llama-cli +COPY --from=build /app/build/ggml/src/libggml.so /libggml.so +COPY --from=build /app/build/src/libllama.so /libllama.so +COPY --from=build /app/build/bin/llama-cli /llama-cli ENTRYPOINT [ "/llama-cli" ] diff --git a/.devops/llama-server-cuda.Dockerfile b/.devops/llama-server-cuda.Dockerfile index 184248984..a40e24205 100644 --- a/.devops/llama-server-cuda.Dockerfile +++ b/.devops/llama-server-cuda.Dockerfile @@ -1,6 +1,6 @@ ARG UBUNTU_VERSION=22.04 # This needs to generally match the container host's environment. -ARG CUDA_VERSION=11.7.1 +ARG CUDA_VERSION=12.6.0 # Target the CUDA build image ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION} # Target the CUDA runtime image @@ -8,33 +8,34 @@ ARG BASE_CUDA_RUN_CONTAINER=nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_V FROM ${BASE_CUDA_DEV_CONTAINER} AS build -# Unless otherwise specified, we make a fat build. -ARG CUDA_DOCKER_ARCH=all +# CUDA architecture to build for (defaults to all supported archs) +ARG CUDA_DOCKER_ARCH=default RUN apt-get update && \ - apt-get install -y build-essential git libcurl4-openssl-dev + apt-get install -y build-essential git cmake libcurl4-openssl-dev WORKDIR /app COPY . . -# Set nvcc architecture -ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH} -# Enable CUDA -ENV GGML_CUDA=1 -# Enable cURL -ENV LLAMA_CURL=1 -# Must be set to 0.0.0.0 so it can listen to requests from host machine -ENV LLAMA_ARG_HOST=0.0.0.0 - -RUN make -j$(nproc) llama-server +# Use the default CUDA archs if not specified +RUN if [ "${CUDA_DOCKER_ARCH}" != "default" ]; then \ + export CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CUDA_DOCKER_ARCH}"; \ + fi && \ + cmake -B build -DGGML_CUDA=ON -DLLAMA_CURL=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \ + cmake --build build --config Release --target llama-server -j$(nproc) FROM ${BASE_CUDA_RUN_CONTAINER} AS runtime RUN apt-get update && \ apt-get install -y libcurl4-openssl-dev libgomp1 curl -COPY --from=build /app/llama-server /llama-server +COPY --from=build /app/build/ggml/src/libggml.so /libggml.so +COPY --from=build /app/build/src/libllama.so /libllama.so +COPY --from=build /app/build/bin/llama-server /llama-server + +# Must be set to 0.0.0.0 so it can listen to requests from host machine +ENV LLAMA_ARG_HOST=0.0.0.0 HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ] diff --git a/docs/docker.md b/docs/docker.md index d8922d77d..e25838255 100644 --- a/docs/docker.md +++ b/docs/docker.md @@ -66,8 +66,8 @@ You may want to pass in some different `ARGS`, depending on the CUDA environment The defaults are: -- `CUDA_VERSION` set to `11.7.1` -- `CUDA_DOCKER_ARCH` set to `all` +- `CUDA_VERSION` set to `12.6.0` +- `CUDA_DOCKER_ARCH` set to the cmake build default, which includes all the supported architectures The resulting images, are essentially the same as the non-CUDA images: