Update Dockerfile

Dockerfile

@@ -1,34 +1,25 @@
-# Use a lightweight Python base
+# Use a lightweight Python base (Debian-based)
 FROM python:3.10-slim
 
 # Set working directory
 WORKDIR /app
 
-# 1. Install HEAVY Build Dependencies
-# We include 'cmake', 'ninja-build', and 'pkg-config' to ensure
-# llama-cpp-python can compile from source if it can't find a wheel.
+# 1. Install System Dependencies
+# 'cmake' and 'build-essential' are strictly required if wheel installation fails.
 RUN apt-get update && apt-get install -y \
     git \
     curl \
     build-essential \
     cmake \
-    ninja-build \
-    pkg-config \
-    gcc \
-    g++ \
+    libopenblas-dev \
     && rm -rf /var/lib/apt/lists/*
 
-# 2. Upgrade pip to the latest version (Crucial for modern wheels)
-RUN pip install --upgrade pip setuptools wheel
-
-# 3. Download Retro Font (VT323)
+# 2. Download Retro Font (VT323)
 RUN curl -L -o /app/VT323.ttf https://github.com/google/fonts/raw/main/ofl/vt323/VT323-Regular.ttf
 
-# 4. Install Dependencies - VERBOSE MODE
-# We set CMAKE_ARGS to simplify the build and use '-v' to show progress.
-# This prevents the "silent freeze" by showing you the compilation logs.
-ENV CMAKE_ARGS="-DLLAMA_NATIVE=OFF -DGGML_NATIVE=OFF"
-RUN pip install --no-cache-dir -v \
+# 3. Install Python Dependencies
+# We split the installation to ensure core libs are present before llama-cpp-python attempts to load.
+RUN pip install --no-cache-dir \
     torch \
     torchvision \
     numpy \
@@ -43,19 +34,21 @@ RUN pip install --no-cache-dir -v \
     safetensors \
     scipy
 
-# 5. Install Llama-CPP separately with Force Logic
-# We try to force a binary. If it fails, it falls back to a verbose build.
-RUN pip install --no-cache-dir -v llama-cpp-python==0.2.90 \
-    --prefer-binary \
-    --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
+# 4. Install Llama-CPP-Python (Corrected)
+# REMOVED: The '--prefer-binary' flag pointing to the wrong index.
+# CHANGED: We now let pip build from source using the installed CMake/GCC. 
+# This guarantees compatibility with the Debian container (glibc).
+# We set CMAKE_ARGS to ensure it builds a CPU-only version (no CUDA req).
+ENV CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_NATIVE=OFF"
+RUN pip install --no-cache-dir llama-cpp-python
 
-# 6. Create a non-root user
+# 5. Create a non-root user
 RUN useradd -m -u 1000 user
 USER user
 ENV HOME=/home/user \
     PATH=/home/user/.local/bin:$PATH
 
-# 7. Write the Monolith Application to disk
+# 6. Write the Monolith Application to disk
 COPY --chown=user <<'EOF' app.py
 import sys, os, io, base64, json, pickle, time
 import numpy as np
@@ -67,7 +60,12 @@ from flask import Flask, request, send_file, render_template_string
 from flask_sock import Sock
 from diffusers import StableDiffusionPipeline, AutoencoderTiny, LCMScheduler
 from PIL import Image, ImageDraw
-from llama_cpp import Llama
+# Wrap Llama import to prevent crash if model not found
+try:
+    from llama_cpp import Llama
+except ImportError:
+    Llama = None
+    print("[!] Warning: llama-cpp-python not loaded correctly.")
 
 # ============================================================================
 # 1. FRONTEND ASSET
@@ -508,6 +506,6 @@ if __name__ == '__main__':
     app.run(host='0.0.0.0', port=7860, threaded=True)
 EOF
 
-# 8. Launch the Monolith
+# 7. Launch the Monolith
 EXPOSE 7860
 CMD ["python", "app.py"]