Update app.py

app.py

@@ -34,7 +34,7 @@ def load_model():
 
 
 @spaces.GPU(duration=120)  # Increased timeout to 120 seconds
-def process_image_and_text(image_pil, text_input, num_beams):
+def process_image_and_text(image_pil, text_input, num_beams, temperature, seed):
     """Extract text from image using PaliGemma2."""
     try:
         processor, model = load_model()
@@ -44,16 +44,18 @@ def process_image_and_text(image_pil, text_input, num_beams):
         image = load_image(image_pil)
 
         # Add <image> token to the beginning of the text prompt
-        text_input = " "+text_input
+        text_input = "<image> " + text_input
 
         # Use the provided text input
         model_inputs = processor(text=text_input, images=image, return_tensors="pt").to(
             device, dtype=torch.bfloat16
         )
         input_len = model_inputs["input_ids"].shape[-1]
+        
+        torch.manual_seed(seed) # Set random seed for reproducibility
 
         with torch.inference_mode():
-            generation = model.generate(**model_inputs, max_new_tokens=200, do_sample=False, num_beams=num_beams)
+            generation = model.generate(**model_inputs, max_new_tokens=100, do_sample=True, num_beams=num_beams, temperature=temperature)
             generation = generation[0][input_len:]
             decoded = processor.decode(generation, skip_special_tokens=True)
 
@@ -70,6 +72,8 @@ if __name__ == "__main__":
             gr.Image(type="pil", label="Upload an image"),
             gr.Textbox(label="Enter Text Prompt"),
             gr.Slider(minimum=1, maximum=10, step=1, value=1, label="Number of Beams"),
+            gr.Slider(minimum=0.1, maximum=2.0, step=0.1, value=1.0, label="Temperature"),
+            gr.Number(label="Random Seed", value=42, precision=0),
         ],
         outputs=gr.Textbox(label="Generated Text"),
         title="PaliGemma2 Image and Text to Text",