refactor(dataset): 重构数据处理脚本并支持自定义数据集

chinese-cosmopedia.py

@@ -0,0 +1,14 @@
+from utils import process_dataset
+
+if __name__ == "__main__":
+    max_chunk_num = 10
+    chunk_size = 1000000
+    item_size = max_chunk_num * chunk_size
+    
+    process_dataset(
+        dataset_name="chinese-cosmopedia",
+        output_subdir="chinese-wiki",
+        split=f"train[:{item_size}]",
+        max_chunk_num=max_chunk_num,
+        chunk_size=chunk_size,
+    )

chinese-wiki.py

@@ -1,8 +0,0 @@
-from utils import process_dataset
-
-if __name__ == "__main__":
-    process_dataset(
-        dataset_name="Blaze7451/Wiki-zh-20250601",
-        output_subdir="chinese-wiki",
-        max_chunk_size=5,
-    )

english-fineweb.py

@@ -1,6 +1,5 @@
 from utils import process_dataset
 
-
 if __name__ == "__main__":
     process_dataset(
         dataset_name="HuggingFaceFW/fineweb",

tokenizer.py

@@ -93,8 +93,8 @@ class BpeTokenizer:
             else:
                 return [encoding.tokens for encoding in encodings]
 
-    def decode(self, tokens: List[int]) -> str:
-        return self._tokenizer.decode(tokens)
+    def decode(self, tokens: List[int], skip_special_tokens=True) -> str:
+        return self._tokenizer.decode(tokens, skip_special_tokens=skip_special_tokens)
     
     def __len__(self) -> int:
         return self._tokenizer.get_vocab_size()

utils.py

@@ -61,19 +61,24 @@ def process_dataset(
     dataset_name: str,
     output_subdir: str,
     dataset_config: str = None,
-    max_chunk_size: int = None,
+    max_chunk_num: int = None,
+    split: str = None,
     split_name: str = "train",
     column_name: str = "text",
     chunk_size: int = 1000000,
     normalization_func=comprehensive_normalization
 ):
 
-    dataset_dict = load_dataset(dataset_name, dataset_config)
-    train_dataset = dataset_dict[split_name]
+    dataset_dict = load_dataset(
+        data_dir=dataset_name, 
+        data_files=dataset_config,
+        split=split
+    )
     
+    train_dataset = dataset_dict[split_name]
     total_samples = len(train_dataset)
     num_chunks = (total_samples // chunk_size) + 1
-    lim_chunks = min(max_chunk_size, num_chunks) if max_chunk_size else num_chunks
+    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)