push

app.py

@@ -54,6 +54,9 @@ async def load_model():
     """Load the Zephyr tokenizer/model and OCR reader on startup"""
     global tokenizer, model, ocr_reader
     try:
+        if tokenizer is not None and model is not None:
+            logger.info("Model already loaded, skipping startup initialization")
+            return
         logger.info(f"Loading model: {MODEL_ID} ...")
         logger.info("Optimizing for CPU-only inference...")