Fix: Force manylinux wheel-2

Dockerfile

@@ -26,26 +26,24 @@ RUN mkdir -p /app/models && chmod 777 /app/models
 # Copy Python requirements first for caching
 COPY requirements.txt .
 
-# Install dependencies
-# Install dependencies
+# Install dependencies (pip + Python deps)
 RUN pip install --no-cache-dir --upgrade pip && \
     pip install --no-cache-dir -r requirements.txt
 
-# Install llama-cpp-python using specific manylinux wheel to avoid musl/glibc mismatch
-RUN pip install https://github.com/abetlen/llama-cpp-python/releases/download/v0.2.90/llama_cpp_python-0.2.90-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+# ⬇️ FIX: install llama-cpp-python correctly (compiles from source)
+RUN CMAKE_ARGS="-DLLAMA_CUBLAS=off" pip install --no-cache-dir llama-cpp-python==0.2.90
 
 # Copy the rest of the application
 COPY . .
 
-# Install the package itself to ensure all dependencies (e.g. cryptography) from pyproject.toml are installed
+# Install your backend package (pyproject.toml)
 RUN pip install .
 
-# Copy built frontend assets from Stage 1
+# Copy built frontend assets
 COPY --from=build-frontend /app/web_client/dist /app/web_client/dist
 
-# Expose the port (Hugging Face expects 7860 by default, but we can set PORT env)
+# Hugging Face defaults
 ENV PORT=7860
 EXPOSE 7860
 
-# Command to run the application
 CMD ["python", "server.py"]