yep

Dockerfile

@@ -1,17 +1,27 @@
 # Use a lightweight Python image
 FROM python:3.10-slim
 
-# 1. Install System Dependencies
-# 'libgl1' replaces 'libgl1-mesa-glx'
-# 'libglib2.0-0' is often needed for cv2/vision
+# 1. Install System Dependencies (Compilers + PDF tools)
+# - build-essential: contains gcc, g++, make (REQUIRED for llama-cpp-python)
+# - cmake: required to build the build system
+# - poppler-utils: required for pdf2image
+# - libgl1: required for opencv/vision tasks
 RUN apt-get update && apt-get install -y \
+    build-essential \
+    cmake \
+    g++ \
+    gcc \
     poppler-utils \
     libgl1 \
     libglib2.0-0 \
     && rm -rf /var/lib/apt/lists/*
 
-# 2. Install Python Libraries
-RUN pip install --no-cache-dir \
+# 2. Upgrade pip (good practice)
+RUN pip install --upgrade pip
+
+# 3. Install Python Libraries
+# We force a reinstall with --no-cache-dir to avoid using broken cached wheels
+RUN CMAKE_ARGS="-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS" pip install --no-cache-dir \
     "llama-cpp-python[server]" \
     huggingface_hub \
     pdf2image \
@@ -20,13 +30,14 @@ RUN pip install --no-cache-dir \
     fastapi \
     pillow
 
-# 3. Download the Model during build
+# 4. Download the Model
+# Using the GGUF (CPU-optimized) version
 RUN huggingface-cli download Qwen/Qwen2-VL-7B-Instruct-GGUF qwen2-vl-7b-instruct-q4_k_m.gguf --local-dir /app/model
 RUN huggingface-cli download Qwen/Qwen2-VL-7B-Instruct-GGUF mmproj-model-f16.gguf --local-dir /app/model
 
-# 4. Copy your API code
+# 5. Copy your API code
 WORKDIR /app
 COPY app.py .
 
-# 5. Run the API Server
+# 6. Run the API Server
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]