Upload paddleocr-vl-1.5.py with huggingface_hub

paddleocr-vl-1.5.py

@@ -412,6 +412,9 @@ def main(
 
     # Load processor and model (native transformers 5.0+ support)
     processor = AutoProcessor.from_pretrained(MODEL_ID)
+    # Set left padding for correct batch generation with decoder-only models
+    processor.tokenizer.padding_side = "left"
+
     model = (
         AutoModelForImageTextToText.from_pretrained(
             MODEL_ID,