compile llama for the docker

Dockerfile

@@ -2,7 +2,22 @@ FROM runpod/pytorch:1.5.0-cu1320-torch2130-ubuntu2404
 WORKDIR /root
 
 RUN mkdir llama.cpp
-COPY llama.cpp ./llama.cpp/
+COPY --exclude=build/ llama.cpp ./llama.cpp/
+WORKDIR /root/llama.cpp
+ENV LD_LIBRARY_PATH=/usr/local/cuda-13.2/compat:${LD_LIBRARY_PATH}
+ENV LIBRARY_PATH=/usr/local/cuda-13.2/compat:${LIBRARY_PATH}
+
+RUN cmake -B build -S . \
+     -DCMAKE_BUILD_TYPE=Release \
+     -DGGML_CUDA=ON \
+     -DCMAKE_CUDA_ARCHITECTURES=120 \
+     -DLLAMA_BUILD_TESTS=OFF \
+     -DLLAMA_BUILD_EXAMPLES=ON \
+    && cmake --build build -j$(nproc) \
+    && cmake --build build --target install \
+    && rm -rf build
+
+WORKDIR /root
 RUN git clone https://github.com/silveroxides/convert_to_quant && cd convert_to_quant && pip install -r requirements.txt && pip install .
 
 RUN mkdir nvfp4_gguf_quant