docker(cuda_simple): upgrade to CUDA 12.8.1 and switch to source install

- Upgrade base image to nvidia/cuda:12.8.1-cudnn-devel-ubuntu22.04
- Optimize deps
- Switch to git clone + pip install . for source building
- Add documentation comments

Signed-off-by: JamePeng <jame_peng@sina.com>

Author: JamePeng

docker/cuda_simple/Dockerfile

@@ -1,27 +1,45 @@
-ARG CUDA_IMAGE="12.5.0-devel-ubuntu22.04"
+# Define the base image version with CUDA 12.8 and cuDNN for Ubuntu 22.04
+ARG CUDA_IMAGE="12.8.1-cudnn-devel-ubuntu22.04"
 FROM nvidia/cuda:${CUDA_IMAGE}
 
 # We need to set the host to 0.0.0.0 to allow outside access
 ENV HOST 0.0.0.0
 
+# Install system dependencies:
 RUN apt-get update && apt-get upgrade -y \
-    && apt-get install -y git build-essential \
-    python3 python3-pip gcc wget \
-    ocl-icd-opencl-dev opencl-headers clinfo \
-    libclblast-dev libopenblas-dev \
-    && mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd
+    && apt-get install -y build-essential \
+    ccache cmake curl gcc git wget \
+    python3 python3-pip python3-wheel \
+    libgomp1 libjpeg-dev libssl-dev libcurl4-openssl-dev
 
+# Set the working directory for the container
+WORKDIR /app
+
+# Copy the current directory contents into the container (useful for local config files or models)
 COPY . .
 
-# setting build related env vars
-ENV CUDA_DOCKER_ARCH=all
+# Set the target GPU architecture (default allows CMake to auto-detect or use common archs)
+ENV CUDA_DOCKER_ARCH=default
 ENV GGML_CUDA=1
+ENV LD_LIBRARY_PATH="/usr/local/cuda/lib64:/usr/local/cuda/compat:/usr/lib/x86_64-linux-gnu:${LD_LIBRARY_PATH}"
+# Enable verbose build output
+ENV VERBOSE=1
+# Specific CUDA paths to ensure the compiler (nvcc) is found
+ENV CUDA_HOME="/usr/local/cuda/"
+ENV CUDA_PATH="/usr/local/cuda/:${PATH}"
+ENV CUDA_TOOLKIT_ROOT_DIR="/usr/local/cuda/"
+
+# Install Python build tools and server dependencies
+RUN python3 -m pip install --upgrade pip pytest wheel packaging scikit-build setuptools fastapi uvicorn pydantic-settings sse-starlette starlette-context huggingface-hub
+
+# Clone the source code. --recursive is CRITICAL to fetch the 'llama.cpp' submodule
+RUN git clone --recursive https://github.com/JamePeng/llama-cpp-python.git
 
-# Install depencencies
-RUN python3 -m pip install --upgrade pip pytest cmake scikit-build setuptools fastapi uvicorn sse-starlette pydantic-settings starlette-context
+# Switch context to the cloned repository directory
+WORKDIR /app/llama-cpp-python
 
 # Install llama-cpp-python (build with cuda)
-RUN CMAKE_ARGS="-DGGML_CUDA=on" pip install llama-cpp-python
+RUN CMAKE_ARGS="-DGGML_CUDA=on -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DCMAKE_CUDA_ARCHITECTURES=${CUDA_DOCKER_ARCH}" pip install .
 
 # Run the server
 CMD python3 -m llama_cpp.server