Update Dockerfile

Dockerfile

@@ -13,4 +13,15 @@ COPY --chown=user ./requirements.txt requirements.txt
 RUN pip install --no-cache-dir --upgrade -r requirements.txt
 
 COPY --chown=user . /app
+
+RUN git clone https://github.com/ikawrakow/ik_llama.cpp.git
+WORKDIR /app/ik_llama.cpp/
+RUN cmake -B build -DGGML_NATIVE=1
+RUN cmake --build build --config Release
+WORKDIR /app/ik_llama.cpp/build/
+RUN wget https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf/resolve/9f43072f69492cbd5bc5d5ebb085fec519686a93/ggml-model-i2_s.gguf
+RUN ./bin/llama-quantize --allow-requantize ./ggml-model-i2_s.gguf model-out.gguf iq2_bn
+
+WORKDIR /app
+
 CMD ["python", "chat_demo.py"]