refactor(utils): 将 pre_tarin_process 和 sft_process 函数合并为 process_dataset

belle_sft.py

@@ -1,9 +1,9 @@
 from datasets import load_dataset
-from utils import pre_tarin_process
+from utils import process_dataset
 
 if __name__ == "__main__":
     dataset = load_dataset("BelleGroup/train_3.5M_CN")
-    # pre_tarin_process(
+    # process_dataset(
     #     dataset_dict=dataset,
     #     output_subdir="belle_sft",
     #     max_chunk_size=5,

chinese-c4.py

@@ -1,9 +1,9 @@
 from datasets import load_dataset
-from utils import pre_tarin_process
+from utils import process_dataset
 
 if __name__ == "__main__":
     dataset = load_dataset("shjwudp/chinese-c4")
-    pre_tarin_process(
+    process_dataset(
         dataset_dict=dataset,
         output_subdir="chinese-c4"
     )

chinese-cosmopedia.py

@@ -1,5 +1,5 @@
 from datasets import load_dataset
-from utils import pre_tarin_process
+from utils import process_dataset
 
 if __name__ == "__main__":
     max_chunk_num = 10
@@ -10,7 +10,7 @@ if __name__ == "__main__":
         data_files={"train": [f"data/0000{i}.parquet" for i in range(5)]}
     )
     
-    pre_tarin_process(
+    process_dataset(
         dataset_dict=dataset,
         output_subdir="chinese-wiki",
         max_chunk_num=max_chunk_num,

english-fineweb.py

@@ -1,9 +1,9 @@
 from datasets import load_dataset
-from utils import pre_tarin_process
+from utils import process_dataset
 
 if __name__ == "__main__":
     dataset = load_dataset("HuggingFaceFW/fineweb", "sample-10BT")
-    pre_tarin_process(
+    process_dataset(
         dataset_dict=dataset,
         output_subdir="english-fineweb",
     )

english-wiki.py

@@ -1,9 +1,9 @@
 from datasets import load_dataset
-from utils import pre_tarin_process
+from utils import process_dataset
 
 if __name__ == "__main__":
     dataset = load_dataset("Blaze7451/enwiki_structured_content")
-    pre_tarin_process(
+    process_dataset(
         dataset_dict=dataset,
         output_subdir="english-wiki",
         max_chunk_size=5,

utils.py

@@ -57,13 +57,14 @@ def dump_pkl_files(
             tensor = torch.cat(arrows)
             pkl.dump(tensor, f)
             
-def pre_tarin_process(
+def process_dataset(
     dataset_dict: DatasetDict,
     output_subdir: str,
     max_chunk_num: int = None,
     chunk_size: int = 1000000,
     split_name: str = "train",
     column_name: str = "text",
+    process_func: Callable[[dict], dict] = None,
     normalization_func=comprehensive_normalization,
 ):
     train_dataset = dataset_dict[split_name]
@@ -83,43 +84,13 @@ def pre_tarin_process(
         output_path = os.path.join(output_dir, f"{output_subdir}_text_chunk_{i}.jsonl")
         with open(output_path, "w", encoding="utf-8") as f:
             for example in chunk:
+                if process_func is not None:
+                    processed_example = process_func(example)
+                else:
                     text = example[column_name]
                     if normalization_func:
                         text = normalization_func(text)
-                json_line = {column_name : text}
+                    processed_example = {column_name: text}
-                f.write(json.dumps(json_line, ensure_ascii=False) + "\n")
+                f.write(json.dumps(processed_example, ensure_ascii=False) + "\n")
-
-        print(f"Saved text chunk {i} to {output_path}")        
-
-
-def sft_process(
-    dataset_dict: DatasetDict,
-    output_subdir: str,
-    max_chunk_num: int = None,
-    chunk_size: int = 1000000,
-    split_name: str = "train",
-    processsor: Callable[[str], str] = None,
-):
-    train_dataset = dataset_dict[split_name]
-    total_samples = len(train_dataset)
-    num_chunks = (total_samples // chunk_size) + 1
-    lim_chunks = min(max_chunk_num, num_chunks) if max_chunk_num else num_chunks
-
-    script_dir = os.path.dirname(os.path.abspath(__file__))
-    output_dir = os.path.join(script_dir, "dataset", output_subdir)
-    os.makedirs(output_dir, exist_ok=True)
-    
-    
-    for i in range(lim_chunks):
-        start_idx = i * chunk_size
-        end_idx = min((i + 1) * chunk_size, total_samples)
-        chunk = train_dataset.select(range(start_idx, end_idx))
-        
-        output_path = os.path.join(output_dir, f"{output_subdir}_text_chunk_{i}.jsonl")
-        with open(output_path, "w", encoding="utf-8") as f:
-            for example in chunk:
-                if processsor is not None:
-                    example = processsor(example)
-                f.write(json.dumps(example, ensure_ascii=False) + "\n")
 
         print(f"Saved text chunk {i} to {output_path}")