Update app.py

app.py

@@ -15,7 +15,7 @@ import cv2
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
-    PaddleOCRVLForConditionalGeneration, # Added for PaddleOCR-VL
+    AutoModelForImageTextToText, # Added for PaddleOCR-VL
     AutoProcessor,
     TextIteratorStreamer,
 )
@@ -136,11 +136,12 @@ model_v = Qwen2_5_VLForConditionalGeneration.from_pretrained(
 MODEL_ID_P = "PaddlePaddle/PaddleOCR-VL"
 SUBFOLDER_P = "PaddleOCR-VL-0.9B"
 processor_p = AutoProcessor.from_pretrained(MODEL_ID_P, trust_remote_code=True, subfolder=SUBFOLDER_P)
-model_p = PaddleOCRVLForConditionalGeneration.from_pretrained(
+model_p = AutoModelForImageTextToText.from_pretrained(
     MODEL_ID_P,
     trust_remote_code=True,
     subfolder=SUBFOLDER_P,
-    torch_dtype=torch.float16
+    torch_dtype=torch.float16,
+    attn_implementation="flash_attention_2"
 ).to(device).eval()