Update Dockerfile

Dockerfile

@@ -9,7 +9,6 @@ RUN apt-get update && \
     wget \
     curl \
     cmake \
-    # python build dependencies \
     build-essential \
     libssl-dev \
     zlib1g-dev \
@@ -23,7 +22,7 @@ RUN apt-get update && \
     libxmlsec1-dev \
     libffi-dev \
     liblzma-dev \
-    ffmpeg \ 
+    ffmpeg \
     libcurl4-openssl-dev \
     nvidia-driver-570
 
@@ -44,10 +43,22 @@ RUN pyenv install ${PYTHON_VERSION} && \
     pip install --no-cache-dir -U pip setuptools wheel && \
     pip install "huggingface-hub" "hf-transfer" "gradio[oauth]>=4.28.0" "gradio_huggingfacehub_search==0.0.8" "APScheduler"
 
-COPY --chown=1000 . /app
+# Clone llama.cpp and install its requirements
 RUN git clone https://github.com/ggerganov/llama.cpp /app/llama.cpp
 RUN pip install -r /app/llama.cpp/requirements.txt
 
+# Build llama.cpp
+ARG RUN_LOCALLY=false
+ENV GGML_CUDA=ON
+RUN if [ "$RUN_LOCALLY" = "true" ]; then export GGML_CUDA=OFF; fi && \
+    echo "GGML_CUDA=$GGML_CUDA" && \
+    cd /app/llama.cpp && \
+    cmake -B build -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=${GGML_CUDA} && \
+    cmake --build build --config Release -j2 --target llama-quantize llama-gguf-split llama-imatrix && \
+    cp ./build/bin/llama-* . && \
+    rm -rf build
+
+COPY --chown=1000 . /app
 COPY groups_merged.txt /app/llama.cpp/
 
 ENV PYTHONPATH=/app \