| |
| """ |
| 修复多图数据集，为第一个用户消息添加 <image> token |
| """ |
|
|
| import json |
| import os |
| from pathlib import Path |
|
|
| def fix_multi_image_data(input_file, output_file): |
| """ |
| 修复多图数据集，在第一个用户消息前添加相应数量的 <image> token |
| |
| Args: |
| input_file: 原始数据文件路径 |
| output_file: 输出文件路径 |
| """ |
| print(f"正在处理文件: {input_file}") |
| |
| |
| with open(input_file, 'r', encoding='utf-8') as f: |
| data = json.load(f) |
| |
| print(f"总样本数: {len(data)}") |
| |
| |
| for i, sample in enumerate(data): |
| |
| num_images = len(sample['images']) |
| |
| |
| image_tokens = '<image>' * num_images |
| |
| |
| conversations = sample['conversations'] |
| for j, conv in enumerate(conversations): |
| if conv['from'] == 'human': |
| |
| original_value = conv['value'] |
| conv['value'] = f"{image_tokens}\n{original_value}" |
| break |
| |
| |
| if (i + 1) % 1000 == 0: |
| print(f"已处理: {i + 1}/{len(data)} 样本") |
| |
| |
| print(f"保存到: {output_file}") |
| with open(output_file, 'w', encoding='utf-8') as f: |
| json.dump(data, f, indent=2, ensure_ascii=False) |
| |
| print("处理完成!") |
| |
| |
| print("\n处理后的示例:") |
| sample = data[0] |
| print(f"图片数量: {len(sample['images'])}") |
| print(f"第一个用户消息: {sample['conversations'][0]['value'][:100]}...") |
|
|
| def main(): |
| |
| base_dir = Path(__file__).parent |
| |
| |
| train_input = base_dir / "data_train_multi_llava_vasevl_v6.json" |
| train_output = base_dir / "data_train_multi_llava_vasevl_v6_fixed.json" |
| |
| |
| test_input = base_dir / "data_test_multi_llava_vasevl_v6.json" |
| test_output = base_dir / "data_test_multi_llava_vasevl_v6_fixed.json" |
| |
| print("开始处理多图数据集...") |
| print("=" * 50) |
| |
| |
| if train_input.exists(): |
| fix_multi_image_data(train_input, train_output) |
| print() |
| else: |
| print(f"训练集文件不存在: {train_input}") |
| |
| |
| if test_input.exists(): |
| fix_multi_image_data(test_input, test_output) |
| else: |
| print(f"测试集文件不存在: {test_input}") |
| |
| print("=" * 50) |
| print("所有文件处理完成!") |
|
|
| if __name__ == "__main__": |
| main() |
|
|
