Change Dockerfile and start.sh to move building llama.cpp to the docker building process. Therefore, we can save time in dev mode to restart the app

start.sh

@@ -1,22 +1,4 @@
 #!/bin/bash
 
-if [ ! -d "llama.cpp" ]; then
-  # only run in dev env
-  git clone https://github.com/ggerganov/llama.cpp
-fi
-
-export GGML_CUDA=OFF
-if [[ -z "${RUN_LOCALLY}" ]]; then
-  # enable CUDA if NOT running locally
-  export GGML_CUDA=ON
-fi
-echo "GGML_CUDA=$GGML_CUDA"
-
-cd llama.cpp
-cmake -B build -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=${GGML_CUDA}
-cmake --build build --config Release -j --target llama-quantize llama-gguf-split llama-imatrix
-cp ./build/bin/llama-* .
-rm -rf build
-
-cd ..
+cd /app
 python app.py