feat(pre_train): 更新 Chinese Cosmopedia 数据集处理脚本

2025-08-06 09:58:15 +08:00
parent d9ef711e28
commit f7200ee4cf
1 changed files with 1 additions and 3 deletions
@@ -2,17 +2,15 @@ from datasets import load_dataset
 from modules.utils import process_dataset

 if __name__ == "__main__":
-    max_chunk_num = 10
    chunk_size = 1000000
    
    dataset = load_dataset(
        "opencsg/chinese-cosmopedia", 
-        data_files={"train": [f"data/0000{i}.parquet" for i in range(5)]}
+        data_files={"train": [f"data/000{i:02d}.parquet" for i in range(25)]}
    )
    
    process_dataset(
        dataset_dict=dataset,
        output_subdir="chinese-wiki-pretrain",
-        max_chunk_num=max_chunk_num,
        chunk_size=chunk_size,
    )