def aggregate_metrics()

in extract_metrics.py [0:0]

• 36 lines of code
• 4 McCabe index (conditional complexity)

def aggregate_metrics(input_folder):
    """Create global_metrics.csv from all subdirectory metrics"""
    top_level_dir = glob.glob(input_folder + '/*')

    for top_dir_path in top_level_dir:
        subdirs = glob.glob(top_dir_path + '/*')    

        aggregated_data = []

        for subdir_path in subdirs:
            metrics_file = os.path.join(subdir_path, 'metrics.csv')
            status_file = os.path.join(subdir_path, 'status.txt')

            folder_name = os.path.basename(subdir_path)

            data = {
                'run_name': folder_name,
                'status': read_status(status_file),
                **parse_folder_name(folder_name)  # Unpack the parsed parameters
            }

            # If metrics.csv exists, read the avg_tokens_s_gpu from it
            if os.path.exists(metrics_file):
                try:
                    with open(metrics_file, 'r') as f:
                        reader = csv.DictReader(f)
                        metrics_data = next(reader)
                        data['avg_tokens_s_gpu'] = int(metrics_data['avg_tokens_s_gpu'])
                        data['avg_mfu'] = int(metrics_data['avg_mfu'])
                except:
                    data['avg_tokens_s_gpu'] = -1
                    data['avg_mfu'] = -1
            else:
                data['avg_tokens_s_gpu'] = -1
                data['avg_mfu'] = -1

            aggregated_data.append(data)
    
        # Write global metrics file
        output_file = os.path.join(top_dir_path, 'global_metrics.csv')
        fieldnames = ['run_name', 'status', 'dp', 'tp', 'pp', 'micro_batch_size', 
                    'grad_acc', 'seq_len', 'avg_tokens_s_gpu', 'avg_mfu']
        
        with open(output_file, 'w', newline='') as f:
            writer = csv.DictWriter(f, fieldnames=fieldnames)
            writer.writeheader()
            writer.writerows(aggregated_data)
            
        print(f"Created global_metrics.csv with {len(aggregated_data)} entries")