deploy

Dockerfile

@@ -1,23 +1,9 @@
-# Use official Python 3.9 image
 FROM python:3.10-slim
-
-# Set working directory
 WORKDIR /app
-
-# Copy requirements file
 COPY requirements.txt .
-
-# Install dependencies
 RUN pip install --no-cache-dir -r requirements.txt
-
-# Create cache directory with proper permissions
+# Create and set permissions for cache directory
 RUN mkdir -p /app/cache && chmod -R 777 /app/cache
-
-# Copy application code
-COPY . .
-
-# Expose port 8000
+COPY app/ .
 EXPOSE 7860
-
-# Start FastAPI with uvicorn
 CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

services/embedding.py

@@ -5,10 +5,15 @@ from typing import List, Optional
 import torch
 import os
 
-# model = SentenceTransformer("clip-ViT-B-32")
-# Set a writable cache directory
+# Set multiple environment variables to redirect all caching
 os.environ["TRANSFORMERS_CACHE"] = "/app/cache"
 os.environ["HF_HOME"] = "/app/cache"
+os.environ["XDG_CACHE_HOME"] = "/app/cache"  # Additional variable for broader compatibility
+
+# Ensure cache directory exists (redundant with Dockerfile but added for safety)
+cache_dir = "/app/cache"
+if not os.path.exists(cache_dir):
+    os.makedirs(cache_dir, exist_ok=True)
 
 # Load model with custom cache directory
 model = SentenceTransformer("clip-ViT-B-32")