try gpt

app.py

@@ -7,17 +7,15 @@ import torch
 from ast import literal_eval
 from PIL import Image
 
-
-# default: Load the model on the available device(s)
+# Load the model on the available device(s)
 model = Qwen2VLForConditionalGeneration.from_pretrained(
     "Qwen/Qwen2-VL-7B-Instruct", torch_dtype="auto", device_map="auto"
 )
 
-# default processer
+# Load the processor
 processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
 
-
-
+# Define your prompts
 other_benifits = '''Extract the following information in the given format:
         {'other_benefits_and_information': {
             '401k eru: {'This Period':'', 'Year-to-Date':''}},
@@ -44,10 +42,10 @@ tax_deductions = '''Extract the following information in the given format:
                 }
 '''
 
-def demo(image_name, prompt):
+def demo(image, prompt):
+    # Convert image to bytes or suitable format if needed
+    image_name = image  # If you're directly using the image
 
-    print("Inside Demo")
-    
     messages = [
         {
             "role": "user",
@@ -61,8 +59,6 @@ def demo(image_name, prompt):
         }
     ]
 
-    print(f"Formulated prompt template {messages}")
-
     # Preparation for inference
     text = processor.apply_chat_template(
         messages, tokenize=False, add_generation_prompt=True
@@ -85,45 +81,33 @@ def demo(image_name, prompt):
     output_text = processor.batch_decode(
         generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
     )
+
+    # Handle output text to convert it into JSON
     try:
-        # almost_json = output_text[0].replace('```\n', '').replace('\n```', '')
-        almost_json = output_text[0].split('```\n')[-1].split('\n```')[0]
- 
+        almost_json = output_text[0].split('\n')[-1].split('\n')[0]
         json = literal_eval(almost_json)
     except:
-        try:
-            # almost_json = output_text[0].replace('```json\n', '').replace('\n```', '')
-            almost_json = output_text[0].split('```json\n')[-1].split('\n```')[0]
-            json = literal_eval(almost_json)
-        except:
-            json = output_text[0]
+        json = output_text[0]  # Return raw output if JSON parsing fails
     return json
 
 def process_document(image):
-    print(f"Received Image --->>>>>> {image}")
-    if isinstance(image, np.ndarray):
-        print("Image is in Numpy array")
-        image = Image.fromarray(image)
-    print(type(image))
-    print("Proceeding with the demo")
+    # Pass the image directly
     one = demo(image, other_benifits)
     two = demo(image, tax_deductions)
     json_op = {
         "tax_deductions": one,
         "other_benifits": two
-              }
+    }
     return json_op
 
-# article = "<p style='text-align: center'><a href='https://www.xelpmoc.in/' target='_blank'>Made by Xelpmoc</a></p>"
-
+# Create Gradio interface
 demo = gr.Interface(
     fn=process_document,
-    inputs=gr.Image(type="pil"),
+    inputs="image",  # Gradio will handle the image input
     outputs="json",
     title="PaySlip_Demo_Model",
-    # article=article,
-    # enable_queue=True,
-    examples=["Slip_1.jpg", "Slip_2.jpg"],
-    cache_examples=False)
+    examples=[["Slip_1.jpg"], ["Slip_2.jpg"]],
+    cache_examples=False
+)
 
-demo.launch()
+demo.launch()