Commit 
							
							·
						
						916ff8e
	
1
								Parent(s):
							
							ea32fa8
								
keep viewer
Browse files- olmocr2-vllm.py +22 -14
    	
        olmocr2-vllm.py
    CHANGED
    
    | @@ -96,7 +96,7 @@ def parse_yaml_frontmatter(text: str) -> tuple[dict, str]: | |
| 96 | 
             
                    (metadata_dict, content_without_frontmatter)
         | 
| 97 | 
             
                """
         | 
| 98 | 
             
                # Match YAML front matter between --- markers
         | 
| 99 | 
            -
                pattern = r | 
| 100 | 
             
                match = re.match(pattern, text.strip(), re.DOTALL)
         | 
| 101 |  | 
| 102 | 
             
                if match:
         | 
| @@ -179,7 +179,6 @@ def create_dataset_card( | |
| 179 | 
             
                model_name = model.split("/")[-1]
         | 
| 180 |  | 
| 181 | 
             
                return f"""---
         | 
| 182 | 
            -
            viewer: false
         | 
| 183 | 
             
            tags:
         | 
| 184 | 
             
            - ocr
         | 
| 185 | 
             
            - document-processing
         | 
| @@ -315,6 +314,7 @@ def main( | |
| 315 | 
             
                    seed: Random seed for shuffling
         | 
| 316 | 
             
                """
         | 
| 317 | 
             
                import time
         | 
|  | |
| 318 | 
             
                start_time = time.time()
         | 
| 319 |  | 
| 320 | 
             
                # Check CUDA availability
         | 
| @@ -396,7 +396,9 @@ def main( | |
| 396 |  | 
| 397 | 
             
                        # Log warning if generation didn't finish naturally
         | 
| 398 | 
             
                        if finish_reason != "stop":
         | 
| 399 | 
            -
                            logger.warning( | 
|  | |
|  | |
| 400 |  | 
| 401 | 
             
                        metadata, content = parse_yaml_frontmatter(response_text)
         | 
| 402 | 
             
                        all_outputs.append(content)
         | 
| @@ -405,25 +407,31 @@ def main( | |
| 405 | 
             
                # Add results to dataset
         | 
| 406 | 
             
                # Check if columns already exist and handle appropriately
         | 
| 407 | 
             
                if output_column in ds.column_names:
         | 
| 408 | 
            -
                    logger.warning( | 
|  | |
|  | |
| 409 | 
             
                    ds = ds.remove_columns([output_column])
         | 
| 410 | 
             
                ds = ds.add_column(output_column, all_outputs)
         | 
| 411 |  | 
| 412 | 
             
                if metadata_column_name in ds.column_names:
         | 
| 413 | 
            -
                    logger.warning( | 
|  | |
|  | |
| 414 | 
             
                    ds = ds.remove_columns([metadata_column_name])
         | 
| 415 | 
             
                ds = ds.add_column(metadata_column_name, all_metadata)
         | 
| 416 |  | 
| 417 | 
             
                # Add inference information
         | 
| 418 | 
            -
                inference_info = json.dumps( | 
| 419 | 
            -
                     | 
| 420 | 
            -
             | 
| 421 | 
            -
             | 
| 422 | 
            -
             | 
| 423 | 
            -
             | 
| 424 | 
            -
             | 
| 425 | 
            -
             | 
| 426 | 
            -
             | 
|  | |
|  | |
| 427 |  | 
| 428 | 
             
                # Handle existing inference_info column
         | 
| 429 | 
             
                if inference_info_column in ds.column_names:
         | 
|  | |
| 96 | 
             
                    (metadata_dict, content_without_frontmatter)
         | 
| 97 | 
             
                """
         | 
| 98 | 
             
                # Match YAML front matter between --- markers
         | 
| 99 | 
            +
                pattern = r"^---\s*\n(.*?)\n---\s*\n(.*)$"
         | 
| 100 | 
             
                match = re.match(pattern, text.strip(), re.DOTALL)
         | 
| 101 |  | 
| 102 | 
             
                if match:
         | 
|  | |
| 179 | 
             
                model_name = model.split("/")[-1]
         | 
| 180 |  | 
| 181 | 
             
                return f"""---
         | 
|  | |
| 182 | 
             
            tags:
         | 
| 183 | 
             
            - ocr
         | 
| 184 | 
             
            - document-processing
         | 
|  | |
| 314 | 
             
                    seed: Random seed for shuffling
         | 
| 315 | 
             
                """
         | 
| 316 | 
             
                import time
         | 
| 317 | 
            +
             | 
| 318 | 
             
                start_time = time.time()
         | 
| 319 |  | 
| 320 | 
             
                # Check CUDA availability
         | 
|  | |
| 396 |  | 
| 397 | 
             
                        # Log warning if generation didn't finish naturally
         | 
| 398 | 
             
                        if finish_reason != "stop":
         | 
| 399 | 
            +
                            logger.warning(
         | 
| 400 | 
            +
                                f"Generation did not finish naturally (reason: {finish_reason}), output may be incomplete"
         | 
| 401 | 
            +
                            )
         | 
| 402 |  | 
| 403 | 
             
                        metadata, content = parse_yaml_frontmatter(response_text)
         | 
| 404 | 
             
                        all_outputs.append(content)
         | 
|  | |
| 407 | 
             
                # Add results to dataset
         | 
| 408 | 
             
                # Check if columns already exist and handle appropriately
         | 
| 409 | 
             
                if output_column in ds.column_names:
         | 
| 410 | 
            +
                    logger.warning(
         | 
| 411 | 
            +
                        f"Column '{output_column}' already exists, it will be overwritten"
         | 
| 412 | 
            +
                    )
         | 
| 413 | 
             
                    ds = ds.remove_columns([output_column])
         | 
| 414 | 
             
                ds = ds.add_column(output_column, all_outputs)
         | 
| 415 |  | 
| 416 | 
             
                if metadata_column_name in ds.column_names:
         | 
| 417 | 
            +
                    logger.warning(
         | 
| 418 | 
            +
                        f"Column '{metadata_column_name}' already exists, it will be overwritten"
         | 
| 419 | 
            +
                    )
         | 
| 420 | 
             
                    ds = ds.remove_columns([metadata_column_name])
         | 
| 421 | 
             
                ds = ds.add_column(metadata_column_name, all_metadata)
         | 
| 422 |  | 
| 423 | 
             
                # Add inference information
         | 
| 424 | 
            +
                inference_info = json.dumps(
         | 
| 425 | 
            +
                    {
         | 
| 426 | 
            +
                        "model": model,
         | 
| 427 | 
            +
                        "script": "olmocr2-vllm.py",
         | 
| 428 | 
            +
                        "version": "1.0.0",
         | 
| 429 | 
            +
                        "timestamp": datetime.now().isoformat(),
         | 
| 430 | 
            +
                        "batch_size": batch_size,
         | 
| 431 | 
            +
                        "max_tokens": max_tokens,
         | 
| 432 | 
            +
                        "temperature": temperature,
         | 
| 433 | 
            +
                    }
         | 
| 434 | 
            +
                )
         | 
| 435 |  | 
| 436 | 
             
                # Handle existing inference_info column
         | 
| 437 | 
             
                if inference_info_column in ds.column_names:
         | 

