sparrow-parse 1.0.6__py3-none-any.whl → 1.0.8__py3-none-any.whl

sparrow_parse/__init__.py

@@ -1 +1 @@
-__version__ = '1.0.6'
+__version__ = '1.0.8'

sparrow_parse/vllm/mlx_inference.py

@@ -75,7 +75,6 @@ class MLXInference(ModelInference):
             print(f"Failed to parse JSON: {e}")
             return output_text
 
-
     def load_image_data(self, image_filepath, max_width=1250, max_height=1750):
         """
         Load and resize image while maintaining its aspect ratio.
@@ -155,7 +154,8 @@ class MLXInference(ModelInference):
     def _process_images(self, model, processor, config, file_paths, input_data, apply_annotation):
         """
         Process images and generate responses for each.
-        
+        If apply_annotation=True, don't resize to maintain accurate coordinates.
+
         :param model: The loaded model
         :param processor: The loaded processor
         :param config: Model configuration
@@ -166,33 +166,59 @@ class MLXInference(ModelInference):
         """
         results = []
         for file_path in file_paths:
-            image, width, height = self.load_image_data(file_path)
-            
+            # Load image differently based on annotation requirement
+            if apply_annotation:
+                # For annotation, just load the image without resizing
+                image = load_image(file_path)
+                # We'll skip the resize_shape parameter when generating
+            else:
+                # For non-annotation cases, load with potential resizing
+                image, width, height = self.load_image_data(file_path)
+                # We'll use resize_shape when generating
+
             # Prepare messages based on model type
             messages = self._prepare_messages(input_data, apply_annotation)
-            
+
             # Generate and process response
             prompt = apply_chat_template(processor, config, messages)
-            response, _ = generate(
-                model,
-                processor,
-                prompt,
-                image,
-                resize_shape=(width, height),
-                max_tokens=4000,
-                temperature=0.0,
-                verbose=False
-            )
-            results.append(self.process_response(response))
+
+            if apply_annotation:
+                # When annotation is required, don't use resize_shape
+                # This preserves original coordinate system
+                response, _ = generate(
+                    model,
+                    processor,
+                    prompt,
+                    image,
+                    max_tokens=4000,
+                    temperature=0.0,
+                    verbose=False
+                )
+            else:
+                # For non-annotation cases, use resize_shape for memory efficiency
+                response, _ = generate(
+                    model,
+                    processor,
+                    prompt,
+                    image,
+                    resize_shape=(width, height),
+                    max_tokens=4000,
+                    temperature=0.0,
+                    verbose=False
+                )
+
+            processed_response = self.process_response(response)
+            results.append(processed_response)
             print(f"Inference completed successfully for: {file_path}")
-        
+
         return results
 
 
     def transform_query_with_bbox(self, text_input):
         """
         Transform JSON schema in text_input to include value, bbox, and confidence.
-        Works with both array and object JSON structures.
+        Works with formats like: "retrieve field1, field2. return response in JSON format,
+        by strictly following this JSON schema: [{...}]."
 
         Args:
             text_input (str): The input text containing a JSON schema
@@ -200,38 +226,33 @@ class MLXInference(ModelInference):
         Returns:
             str: Text with transformed JSON including value, bbox, and confidence
         """
-        # Split text into parts - find the JSON portion between "retrieve" and "return response"
-        retrieve_pattern = r'retrieve\s+'
-        return_pattern = r'\.\s+return\s+response'
 
-        retrieve_match = re.search(retrieve_pattern, text_input)
-        return_match = re.search(return_pattern, text_input)
+        schema_pattern = r'JSON schema:\s*(\[.*?\]|\{.*?\})'
+        schema_match = re.search(schema_pattern, text_input, re.DOTALL)
 
-        if not retrieve_match or not return_match:
+        if not schema_match:
             return text_input  # Return original if pattern not found
 
-        json_start = retrieve_match.end()
-        json_end = return_match.start()
-
-        prefix = text_input[:json_start]
-        json_str = text_input[json_start:json_end].strip()
-        suffix = text_input[json_end:]
+        # Extract the schema part and its position
+        schema_str = schema_match.group(1).strip()
+        schema_start = schema_match.start(1)
+        schema_end = schema_match.end(1)
 
         # Parse and transform the JSON
         try:
             # Handle single quotes if needed
-            json_str = json_str.replace("'", '"')
+            schema_str = schema_str.replace("'", '"')
 
-            json_obj = json.loads(json_str)
+            json_obj = json.loads(schema_str)
             transformed_json = self.transform_query_structure(json_obj)
             transformed_json_str = json.dumps(transformed_json)
 
-            # Rebuild the text
-            result = prefix + transformed_json_str + suffix
+            # Rebuild the text by replacing just the schema portion
+            result = text_input[:schema_start] + transformed_json_str + text_input[schema_end:]
 
             return result
         except json.JSONDecodeError as e:
-            print(f"Error parsing JSON: {e}")
+            print(f"Error parsing JSON schema: {e}")
             return text_input  # Return original if parsing fails
 
 

{sparrow_parse-1.0.6.dist-info → sparrow_parse-1.0.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sparrow-parse
-Version: 1.0.6
+Version: 1.0.8
 Summary: Sparrow Parse is a Python package (part of Sparrow) for parsing and extracting information from documents.
 Home-page: https://github.com/katanaml/sparrow/tree/main/sparrow-data/parse
 Author: Andrej Baranovskij

{sparrow_parse-1.0.6.dist-info → sparrow_parse-1.0.8.dist-info}/RECORD

@@ -1,4 +1,4 @@
-sparrow_parse/__init__.py,sha256=zrUEHc9dmvLJ5ka5maZk9TTHoZ21dwKsENXeOSwXM3o,21
+sparrow_parse/__init__.py,sha256=iCEPnhz-knfGRAO4Ep2uQaYf4xwhPIjjcgAcNjga8kc,21
 sparrow_parse/__main__.py,sha256=Xs1bpJV0n08KWOoQE34FBYn6EBXZA9HIYJKrE4ZdG78,153
 sparrow_parse/text_extraction.py,sha256=uhYVNK5Q2FZnw1Poa3JWjtN-aEL7cyKpvaltdn0m2II,8948
 sparrow_parse/extractors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -13,9 +13,9 @@ sparrow_parse/vllm/huggingface_inference.py,sha256=RqYmP-wh_cm_BZ271HbejnZe30S5E
 sparrow_parse/vllm/inference_base.py,sha256=AmWF1OUjJLxSEK_WCbcRpXHX3cKk8nPJJHha_X-9Gs4,844
 sparrow_parse/vllm/inference_factory.py,sha256=FTM65O-dW2WZchHOrNN7_Q3-FlVoAc65iSptuuUuClM,1166
 sparrow_parse/vllm/local_gpu_inference.py,sha256=SIyprv12fYawwfxgQ7ZOTM5WmMfQqhO_9vbereRpZdk,652
-sparrow_parse/vllm/mlx_inference.py,sha256=wNysikBBU5tTg3u2902EkhJOoliccHydL4IXHOW6j3I,11824
-sparrow_parse-1.0.6.dist-info/METADATA,sha256=RKzmkA3uaUQ9g5kJivfYp7S20_gbqeiZ_uJA7_fbmQQ,7229
-sparrow_parse-1.0.6.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
-sparrow_parse-1.0.6.dist-info/entry_points.txt,sha256=HV5nnQVtr2m-kn6hzY_ynp0zugNCcGovbmnfmQgOyhw,53
-sparrow_parse-1.0.6.dist-info/top_level.txt,sha256=n6b-WtT91zKLyCPZTP7wvne8v_yvIahcsz-4sX8I0rY,14
-sparrow_parse-1.0.6.dist-info/RECORD,,
+sparrow_parse/vllm/mlx_inference.py,sha256=j4DWq6e_9iQSt7CmWuA7OD7RoXkCrxzCNq4UffBuaoQ,12882
+sparrow_parse-1.0.8.dist-info/METADATA,sha256=clalm_6WpyInHCLH10dyMGX4dgJrPHIXwSU9ltSFZKM,7229
+sparrow_parse-1.0.8.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
+sparrow_parse-1.0.8.dist-info/entry_points.txt,sha256=HV5nnQVtr2m-kn6hzY_ynp0zugNCcGovbmnfmQgOyhw,53
+sparrow_parse-1.0.8.dist-info/top_level.txt,sha256=n6b-WtT91zKLyCPZTP7wvne8v_yvIahcsz-4sX8I0rY,14
+sparrow_parse-1.0.8.dist-info/RECORD,,