Update Dockerfile

Dockerfile

@@ -1,35 +1,30 @@
-FROM python:3.11
+FROM python:3.11-slim
 
 ENV DEBIAN_FRONTEND=noninteractive \
-    FORCE_CMAKE=1 \
     PYTHONUNBUFFERED=1
 
 WORKDIR /app
 
-# Install build tools and libs needed by llama-cpp-python
+# Minimal tools to download the wheel
 RUN apt-get update && apt-get install -y --no-install-recommends \
-    build-essential \
-    cmake \
-    git \
-    pkg-config \
-    libopenblas-dev \
-    libomp-dev \
-    libgomp1 \
-    python3-dev \
+    ca-certificates \
+    curl \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy only the requirements first so this layer can be cached when app sources change
+# Copy requirements (do NOT include llama-cpp-python here; it's installed separately)
 COPY requirements.txt /app/requirements.txt
 
-# Upgrade pip tooling and install requirements
-RUN pip install --upgrade pip setuptools wheel && \
+# Download the prebuilt wheel and install it, then install other Python deps
+RUN mkdir -p /wheels && \
+    curl -L -o /wheels/llama_cpp_python-0.2.75+cpubasic-cp311-cp311-linux_x86_64.whl \
+    "https://github.com/kuwaai/llama-cpp-python-wheels/releases/download/cpu/llama_cpp_python-0.2.75+cpubasic-cp311-cp311-linux_x86_64.whl" && \
+    pip install --upgrade pip setuptools wheel && \
+    pip install --no-cache-dir /wheels/llama_cpp_python-0.2.75+cpubasic-cp311-cp311-linux_x86_64.whl && \
     pip install --no-cache-dir -r /app/requirements.txt
 
-# Install llama-cpp-python (build from source if wheel not available)
-RUN pip install --no-cache-dir llama-cpp-python
-
-# Copy app sources
+# Copy app sources last to keep previous layers cacheable
 COPY app.py /app/app.py
 
 EXPOSE 7860 7861
-CMD ["python", "app.py"]
+
+CMD ["python", "app.py"]