ok

Dockerfile

@@ -5,41 +5,37 @@ FROM python:3.11-slim
 ENV PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \
     PYTHONIOENCODING=UTF-8 \
-    # Point HF cache to a writable directory inside /app
+    # Set HF_HOME to a writable directory
     HF_HOME=/app/cache \
     TRANSFORMERS_CACHE=/app/cache
 
-# Install system dependencies (curl) and clean up in the same layer to reduce image size
+# Install system dependencies
 RUN apt-get update && apt-get install -y --no-install-recommends curl \
     && rm -rf /var/lib/apt/lists/* \
     && useradd -m -u 1000 user
 
-# Set working directory
 WORKDIR /app
 
-# --- LAYER 1: Dependencies (Cached unless requirements.txt changes) ---
+# --- LAYER 1: Dependencies ---
 COPY --chown=user:user requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# --- LAYER 2: Model Files (Cached unless model weights change) ---
-# We explicitly copy the models folder before the rest of the code.
-# This prevents the heavy model layer from breaking cache when you just edit main.py.
-# Note: Ensure you have a 'models' folder locally. If not, you can remove this line.
-COPY --chown=user:user models ./models
+# --- LAYER 2: Model Download (Cached) ---
+# Instead of copying local files, we download the model during the build.
+# This layer will be CACHED and won't run again unless you change this line.
+RUN python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='BAAI/bge-base-en-v1.5', local_dir='./models/bge-base-en-v1.5')"
 
-# --- LAYER 3: Application Code (Cached until you edit code) ---
+# --- LAYER 3: Application Code ---
 COPY --chown=user:user . .
 
-# Ensure the cache directory exists and the user owns it
-RUN mkdir -p $HF_HOME && chown -R user:user /app/cache
+# Ensure permissions
+RUN mkdir -p $HF_HOME && chown -R user:user /app/cache && chown -R user:user /app/models
 
-# Switch to the non-root user
+# Switch user
 USER user
 
-# Expose port 7860
+# Expose port
 EXPOSE 7860
 
 # Start script
-# 1. Background loop pings the HF Space URL to keep it alive
-# 2. Uvicorn runs the app
 CMD bash -c "while true; do curl -s https://sasasas635-database-chat.hf.space/ping >/dev/null && sleep 300; done & uvicorn main:app --host 0.0.0.0 --port 7860 --workers 4 --loop asyncio"