revert

app.py

@@ -25,6 +25,12 @@ DEFAULT_NER_LABELS = "person, organization, location, date, event"
 
 # }
 
+class TextWithMetadata(list):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args)
+        self.original_text = kwargs.get('original_text', '')
+        self.entities = kwargs.get('entities', [])
+
 def array_to_image_path(image_array):
     # Convert numpy array to PIL Image
     img = Image.fromarray(np.uint8(image_array))
@@ -43,26 +49,14 @@ def array_to_image_path(image_array):
     return full_path
     
 models = {
-    "Qwen/Qwen2.5-VL-7B-Instruct": Qwen2_5_VLForConditionalGeneration.from_pretrained(
-        "Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto"
-    ).cuda().eval(),
+    "Qwen/Qwen2.5-VL-7B-Instruct": Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto").cuda().eval()
 
-    "medieval-data/qwen2.5-vl-old-church-slavonic": Qwen2_5_VLForConditionalGeneration.from_pretrained(
-        "medieval-data/qwen2.5-vl-old-church-slavonic", trust_remote_code=True, torch_dtype="auto"
-    ).cuda().eval()
 }
 
 processors = {
-    "Qwen/Qwen2.5-VL-7B-Instruct": AutoProcessor.from_pretrained(
-        "Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True
-    ),
-
-    "medieval-data/qwen2.5-vl-old-church-slavonic": AutoProcessor.from_pretrained(
-        "medieval-data/qwen2.5-vl-old-church-slavonic", trust_remote_code=True
-    )
+    "Qwen/Qwen2.5-VL-7B-Instruct": AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True)
 }
 
-
 DESCRIPTION = "This demo uses[Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct)"
 
 kwargs = {}
@@ -74,104 +68,92 @@ prompt_suffix = "<|end|>\n"
 
 @spaces.GPU
 def run_example(image, model_id="Qwen/Qwen2.5-VL-7B-Instruct", run_ner=False, ner_labels=DEFAULT_NER_LABELS):
-    try:
-        # First get the OCR text
-        text_input = "Convert the image to text."
-        image_path = array_to_image_path(image)
-        
-        model = models[model_id]
-        processor = processors[model_id]
-        
-        prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
-        image = Image.fromarray(image).convert("RGB")
-        messages = [
-        {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "image",
-                        "image": image_path,
-                    },
-                    {"type": "text", "text": text_input},
-                ],
-            }
-        ]
-        
-        # Preparation for inference
-        text = processor.apply_chat_template(
-            messages, tokenize=False, add_generation_prompt=True
-        )
-        image_inputs, video_inputs = process_vision_info(messages)
-        inputs = processor(
-            text=[text],
-            images=image_inputs,
-            videos=video_inputs,
-            padding=True,
-            return_tensors="pt",
-        )
-        inputs = inputs.to("cuda")
-        
-        # Inference: Generation of the output
-        generated_ids = model.generate(**inputs, max_new_tokens=1024)
-        generated_ids_trimmed = [
-            out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
-        ]
-        output_text = processor.batch_decode(
-            generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    # First get the OCR text
+    text_input = "Convert the image to text."
+    image_path = array_to_image_path(image)
+    
+    model = models[model_id]
+    processor = processors[model_id]
+    
+    prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
+    image = Image.fromarray(image).convert("RGB")
+    messages = [
+    {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image_path,
+                },
+                {"type": "text", "text": text_input},
+            ],
+        }
+    ]
+    
+    # Preparation for inference
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+    
+    # Inference: Generation of the output
+    generated_ids = model.generate(**inputs, max_new_tokens=1024)
+    generated_ids_trimmed = [
+        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    
+    ocr_text = output_text[0]
+    
+    # If NER is enabled, process the OCR text
+    if run_ner:
+        ner_results = gliner_model.predict_entities(
+            ocr_text,
+            ner_labels.split(","),
+            threshold=0.3
         )
         
-        ocr_text = output_text[0]
+        # Create a list of tuples (text, label) for highlighting
+        highlighted_text = []
+        last_end = 0
         
-        # Create state dictionary
-        state_dict = {
-            "original_text": ocr_text,
-            "entities": []
-        }
+        # Sort entities by start position
+        sorted_entities = sorted(ner_results, key=lambda x: x["start"])
         
-        # If NER is enabled, process the OCR text
-        if run_ner:
-            ner_results = gliner_model.predict_entities(
-                ocr_text,
-                ner_labels.split(","),
-                threshold=0.3
-            )
+        # Process each entity and add non-entity text segments
+        for entity in sorted_entities:
+            # Add non-entity text before the current entity
+            if last_end < entity["start"]:
+                highlighted_text.append((ocr_text[last_end:entity["start"]], None))
             
-            # Update state with entities
-            state_dict["entities"] = ner_results
-            
-            # Create a list of tuples (text, label) for highlighting
-            highlighted_text = []
-            last_end = 0
-            
-            # Sort entities by start position
-            sorted_entities = sorted(ner_results, key=lambda x: x["start"])
-            
-            # Process each entity and add non-entity text segments
-            for entity in sorted_entities:
-                # Add non-entity text before the current entity
-                if last_end < entity["start"]:
-                    highlighted_text.append((ocr_text[last_end:entity["start"]], None))
-                
-                # Add the entity text with its label
-                highlighted_text.append((
-                    ocr_text[entity["start"]:entity["end"]],
-                    entity["label"]
-                ))
-                last_end = entity["end"]
-            
-            # Add any remaining text after the last entity
-            if last_end < len(ocr_text):
-                highlighted_text.append((ocr_text[last_end:], None))
-            
-            return highlighted_text, state_dict
+            # Add the entity text with its label
+            highlighted_text.append((
+                ocr_text[entity["start"]:entity["end"]],
+                entity["label"]
+            ))
+            last_end = entity["end"]
         
-        # If NER is disabled, return the text without highlighting
-        highlighted_text = [(ocr_text, None)]
-        return highlighted_text, state_dict
+        # Add any remaining text after the last entity
+        if last_end < len(ocr_text):
+            highlighted_text.append((ocr_text[last_end:], None))
         
-    except Exception as e:
-        error_msg = f"Error processing image: {str(e)}"
-        return [(error_msg, None)], {"original_text": error_msg, "entities": []}
+        # Create TextWithMetadata instance with the highlighted text and metadata
+        result = TextWithMetadata(highlighted_text, original_text=ocr_text, entities=ner_results)
+        return result, result  # Return twice: once for display, once for state
+    
+    # If NER is disabled, return the text without highlighting
+    result = TextWithMetadata([(ocr_text, None)], original_text=ocr_text, entities=[])
+    return result, result  # Return twice: once for display, once for state
 
 css = """
   /* Overall app styling */
@@ -282,7 +264,7 @@ with gr.Blocks(css=css) as demo:
     # Modify create_zip to use the state data
     def create_zip(image, fname, ocr_result):
         # Validate inputs
-        if not fname or image is None:
+        if not fname or image is None:  # Changed the validation check
             return None
         
         try:
@@ -297,9 +279,9 @@ with gr.Blocks(css=css) as demo:
                 img_path = os.path.join(temp_dir, f"{fname}.png")
                 image.save(img_path)
                 
-                # Use the OCR result from state - now it's a dictionary
-                original_text = ocr_result.get("original_text", "") if ocr_result else ""
-                entities = ocr_result.get("entities", []) if ocr_result else []
+                # Use the OCR result from state
+                original_text = ocr_result.original_text if ocr_result else ""
+                entities = ocr_result.entities if ocr_result else []
                 
                 # Save text
                 txt_path = os.path.join(temp_dir, f"{fname}.txt")