diff --git a/pre_train/__init__.py b/pre_train/__init__.py
new file mode 100644
index 0000000..c5a770d
--- /dev/null
+++ b/pre_train/__init__.py
@@ -0,0 +1,12 @@
+import sys
+import os
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import utils
+from tokenizer import BpeTokenizer
+
+__all__ = [
+    "utils",
+    "BpeTokenizer",
+]
\ No newline at end of file
diff --git a/chinese-c4.py b/pre_train/chinese-c4.py
similarity index 100%
rename from chinese-c4.py
rename to pre_train/chinese-c4.py
diff --git a/chinese-cosmopedia.py b/pre_train/chinese-cosmopedia.py
similarity index 100%
rename from chinese-cosmopedia.py
rename to pre_train/chinese-cosmopedia.py
diff --git a/english-fineweb.py b/pre_train/english-fineweb.py
similarity index 100%
rename from english-fineweb.py
rename to pre_train/english-fineweb.py
diff --git a/english-wiki.py b/pre_train/english-wiki.py
similarity index 100%
rename from english-wiki.py
rename to pre_train/english-wiki.py
diff --git a/run.py b/run.py
new file mode 100644
index 0000000..432a48c
--- /dev/null
+++ b/run.py
@@ -0,0 +1,47 @@
+# run_all.py
+
+import os
+import sys
+import importlib.util
+
+# 确保根目录在路径中
+PROJECT_ROOT = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(PROJECT_ROOT)
+
+def run_script(script_path):
+    """动态导入并运行一个 Python 脚本"""
+    if not os.path.exists(script_path):
+        print(f"[警告] 文件不存在: {script_path}")
+        return
+
+    # 生成模块名
+    module_name = os.path.splitext(os.path.basename(script_path))[0]
+    spec = importlib.util.spec_from_file_location(module_name, script_path)
+    module = importlib.util.module_from_spec(spec)
+
+    # 插入到 sys.modules 避免重复导入
+    sys.modules[module_name] = module
+
+    # 执行脚本（相当于 __name__ == "__main__"）
+    print(f"\n{'='*50}")
+    print(f"运行: {script_path}")
+    print(f"{'='*50}")
+    spec.loader.exec_module(module)
+
+def main():
+    # 运行 pre_train 下的所有脚本
+    pre_train_dir = os.path.join(PROJECT_ROOT, 'pre_train')
+    for file in os.listdir(pre_train_dir):
+        if file.endswith('.py') and not file.startswith('__'):
+            script_path = os.path.join(pre_train_dir, file)
+            run_script(script_path)
+
+    # 运行 supervised_finetuning 下的所有脚本
+    sft_dir = os.path.join(PROJECT_ROOT, 'supervised_finetuning')
+    for file in os.listdir(sft_dir):
+        if file.endswith('.py') and not file.startswith('__'):
+            script_path = os.path.join(sft_dir, file)
+            run_script(script_path)
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/supervised_finetuning/__init__.py b/supervised_finetuning/__init__.py
new file mode 100644
index 0000000..c5a770d
--- /dev/null
+++ b/supervised_finetuning/__init__.py
@@ -0,0 +1,12 @@
+import sys
+import os
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import utils
+from tokenizer import BpeTokenizer
+
+__all__ = [
+    "utils",
+    "BpeTokenizer",
+]
\ No newline at end of file
diff --git a/sft_belle.py b/supervised_finetuning/sft_belle.py
similarity index 100%
rename from sft_belle.py
rename to supervised_finetuning/sft_belle.py
diff --git a/sft_chinese_instruct.py b/supervised_finetuning/sft_chinese_instruct.py
similarity index 100%
rename from sft_chinese_instruct.py
rename to supervised_finetuning/sft_chinese_instruct.py
diff --git a/sft_coder.py b/supervised_finetuning/sft_coder.py
similarity index 100%
rename from sft_coder.py
rename to supervised_finetuning/sft_coder.py
diff --git a/sft_magpie-pro-300k.py b/supervised_finetuning/sft_magpie-pro-300k.py
similarity index 100%
rename from sft_magpie-pro-300k.py
rename to supervised_finetuning/sft_magpie-pro-300k.py
diff --git a/utils.py b/utils.py
index 86906b8..9174d24 100644
--- a/utils.py
+++ b/utils.py
@@ -116,14 +116,16 @@ def process_dataset(
     column_name: str = "text",
     process_func: Union[Callable[[dict], dict], Callable[[List[dict]], List[dict]]] = None,
     normalization_func=comprehensive_normalization,
+    output_dir: 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)
+    if output_dir is None:
+        output_dir = os.path.join(os.getcwd(), "dataset", output_subdir)
+        
     os.makedirs(output_dir, exist_ok=True)
 
     for i in range(lim_chunks):