Update app.py

app.py

@@ -23,17 +23,15 @@ feature_extractor= ViTImageProcessor.from_pretrained(model_path)
 
 # Load model
 model = VisionEncoderDecoderModel.from_pretrained(model_path)
-pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values
-# use GPT2's eos_token as the pad as well as eos token
 
 
 
 def predict(image,max_length=24, num_beams=4):
     image = image.convert('RGB')
-    sequences = model.generate(pixel_values, num_beams=4, max_length=25)
+    pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values
     sequences = model.generate(pixel_values, num_beams=4, max_length=25)
     captions = tokenizer.batch_decode(sequences, skip_special_tokens=True)
-  return captions
+    return captions
 
 
 # Gradio Interface