scripts/warmup_models.py

#!/usr/bin/env python3
"""Pre-warm embedding models to avoid first-request timeout"""

print("Pre-warming embedding models...")

try:
    from sentence_transformers import SentenceTransformer, CrossEncoder
    
    print("Loading sentence transformer (all-MiniLM-L6-v2)...")
    SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
    print("✓ Sentence transformer loaded")
    
    print("Loading cross-encoder (ms-marco-MiniLM-L-6-v2)...")
    CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
    print("✓ Cross-encoder loaded")
    
    print("All models cached successfully!")
except Exception as e:
    print(f"Warning: Model warmup failed: {e}")
    print("Models will be downloaded on first request")