use llama-cpp-python wheel from HF dataset

Dockerfile

@@ -1,12 +1,11 @@
-# Use official slim Python image
-FROM python:3.10-slim
+# Use official Python image
+FROM python:3.10
 
 # Install system dependencies
 RUN apt-get update && apt-get install -y \
     git \
     wget \
     build-essential \
-    cmake \
     libopenblas-dev \
     libcurl4-openssl-dev \
     curl \
@@ -18,17 +17,15 @@ WORKDIR /app
 # Copy all project files
 COPY . .
 
-# Clone and build llama.cpp
-RUN git clone https://github.com/ggerganov/llama.cpp.git /app/llama.cpp && \
-    cd /app/llama.cpp && \
-    mkdir -p build && cd build && \
-    cmake .. -DLLAMA_BUILD_EXAMPLES=ON -DLLAMA_BUILD_SERVER=OFF -DLLAMA_CURL=OFF && \
-    cmake --build . --config Release
+# Download and install llama-cpp-python prebuilt wheel from Hugging Face Dataset
+RUN wget https://huggingface.co/datasets/Kalpokoch/wheel-llama/resolve/main/llama_cpp_python-0.3.13-cp311-cp311-linux_x86_64.whl -O llama_cpp_python.whl && \
+    pip install llama_cpp_python.whl && \
+    rm llama_cpp_python.whl
 
-# Install Python dependencies
+# Install remaining Python dependencies
 RUN pip install --no-cache-dir -r requirements.txt
 
-# ✅ Ensure /app/models exists and is writable
+# Ensure /app/models exists and is writable
 RUN mkdir -p /app/models && chmod -R 777 /app/models
 
 # Expose FastAPI port