docker: cleanup Dockerfile.cuda

Dockerfile.cuda

@@ -1,81 +1,68 @@
-FROM registry.access.redhat.com/ubi9/ubi:latest as base
-
-ENV CUDA_VERSION=11.8.0 \
-    NV_CUDA_LIB_VERSION=11.8.0-1 \
-    NVIDIA_VISIBLE_DEVICES=all \
-    NVIDIA_DRIVER_CAPABILITIES=compute,utility \
-    NV_CUDA_CUDART_VERSION=11.8.89-1 \
-    NV_CUDA_COMPAT_VERSION=520.61.05-1
-
-RUN dnf config-manager \
-       --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo \
-    && dnf install -y \
-        cuda-cudart-11-8-${NV_CUDA_CUDART_VERSION} \
-        cuda-compat-11-8-${NV_CUDA_COMPAT_VERSION} \
-    && echo "/usr/local/nvidia/lib" >> /etc/ld.so.conf.d/nvidia.conf \
-    && echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf \
-    && dnf clean all
+FROM registry.access.redhat.com/ubi9/ubi:latest as cuda-base
+ARG CUDA_VERSION=12-3
+
+ENV CUDA_VERSION=${CUDA_VERSION}
+
+RUN dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo && \
+    dnf install -y  \
+        cuda-cudart-${CUDA_VERSION} \
+        cuda-compat-${CUDA_VERSION} \
+        libcublas-${CUDA_VERSION} \
+    && \
+    echo "/usr/local/nvidia/lib" >> /etc/ld.so.conf.d/nvidia.conf && \
+    echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf && \
+    dnf clean all
 
 ENV CUDA_HOME="/usr/local/cuda"
 ENV PATH="/usr/local/nvidia/bin:${CUDA_HOME}/bin:${PATH}"
 ENV LD_LIBRARY_PATH="/usr/local/nvidia/lib:/usr/local/nvidia/lib64:$CUDA_HOME/lib64:$CUDA_HOME/extras/CUPTI/lib64:${LD_LIBRARY_PATH}"
 
 
-FROM base as builder
-ENV NV_NVTX_VERSION=11.8.86-1 \
-    NV_LIBNPP_VERSION=11.8.0.86-1 \
-    NV_LIBCUBLAS_VERSION=11.11.3.6-1 \
-    NV_LIBNCCL_PACKAGE_VERSION=2.15.5-1+cuda11.8 \
-    NVCC_VERSION=11-8 \
-    CUBLAS_VERSION=11-8 \
-    CUDART_VERSION=11-8
+FROM cuda-base as builder
+ARG CUDA_VERSION
+ENV CUDA_VERSION=${CUDA_VERSION}
 
 RUN dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo && \
-    dnf -y update && \
     dnf -y install \
-        git gcc-toolset-12-gcc-c++ gcc-toolset-12 cmake shadow-utils python3.11-pip \
-        cuda-nvcc-${NVCC_VERSION} \
-        libcublas-${CUBLAS_VERSION} \
-        cuda-libraries-11-8-${NV_CUDA_LIB_VERSION} \
-        cuda-nvtx-11-8-${NV_NVTX_VERSION} \
-        libnpp-11-8-${NV_LIBNPP_VERSION} \
-        libcublas-11-8-${NV_LIBCUBLAS_VERSION} \
-        libnccl-${NV_LIBNCCL_PACKAGE_VERSION} && \
+        git gcc-toolset-12-gcc-c++ gcc-toolset-12 cmake python3.11-pip \
+        cuda-nvcc-${CUDA_VERSION} \
+        libcublas-devel-${CUDA_VERSION} \
+        cuda-cudart-devel-${CUDA_VERSION} \
+        cuda-driver-devel-${CUDA_VERSION} \
+    && \
     dnf clean all && \
     pip3.11 install --no-cache-dir --upgrade pip wheel
 
+ENV POETRY_VIRTUALENVS_IN_PROJECT=1
 
-RUN dnf -y install cuda-cudart-${CUDART_VERSION}
-# ENV POETRY_VIRTUALENVS_IN_PROJECT=1
-# ENV CMAKE_ARGS="-DLLAMA_CUBLAS=on"
+WORKDIR /llama-cpp-python
+COPY pyproject.toml .
+COPY poetry.lock .
 
-# WORKDIR /llama-cpp-python
-# COPY pyproject.toml .
-# COPY poetry.lock .
-# RUN pip3.11 install poetry && \
-#     scl enable gcc-toolset-12 -- poetry install
+ENV CMAKE_ARGS="-DLLAMA_CUBLAS=on"
+RUN pip3.11 install poetry && \
+    scl enable gcc-toolset-12 -- poetry install
 
 
-# FROM base as deploy
+FROM cuda-base as deploy
 
-# RUN dnf -y update && \
-#     dnf -y install \
-#         shadow-utils python3.11 && \
-#     dnf clean all
+RUN dnf -y install \
+        shadow-utils python3.11 && \
+    dnf clean all
 
-# WORKDIR /llama-cpp-python
+WORKDIR /llama-cpp-python
 
-# COPY --from=builder /llama-cpp-python/.venv /llama-cpp-python/.venv
+COPY --from=builder /llama-cpp-python/.venv /llama-cpp-python/.venv
 
-# ENV VIRTUAL_ENV=/llama-cpp-python/.venv
-# ENV PATH="$VIRTUAL_ENV/bin:$PATH"
+ENV VIRTUAL_ENV=/llama-cpp-python/.venv
+ENV PATH="$VIRTUAL_ENV/bin:$PATH"
 
-# RUN groupadd --system llama --gid 1001 && \
-#     adduser --system --uid 1001 --gid 0 --groups llama \
-#     --create-home --home-dir /llama-cpp-python --shell /sbin/nologin \
-#     --comment "Llama user" llama
+RUN groupadd --system llama --gid 1001 && \
+    adduser --system --uid 1001 --gid 0 --groups llama \
+    --create-home --home-dir /llama-cpp-python --shell /sbin/nologin \
+    --comment "Llama user" llama
 
-# USER llama
+USER llama
 
-# # TODO: configuration
-# ENTRYPOINT ["python", "-m", "llama_cpp.server"]
+# TODO: configuration?
+ENTRYPOINT ["python", "-m", "llama_cpp.server"]