frontend/train_page.py

@@ -1,29 +1,9 @@
 import gradio as gr
-import sys
-from pathlib import Path
-sys.path.append(str(Path(__file__).resolve().parent.parent))
-from global_var import get_model, get_tokenizer, get_datasets
 
 def train_page():
     with gr.Blocks() as demo:
-
         gr.Markdown("## 微调")
-        # 获取数据集列表并设置初始值
+        with gr.Row():
-        datasets_list = [str(ds["name"]) for ds in get_datasets().all()]
+            with gr.Column():
-        initial_dataset = datasets_list[0] if datasets_list else None
+                pass
-        dataset_dropdown = gr.Dropdown(
+    return demo
-            choices=datasets_list,
-            value=initial_dataset,  # 设置初始选中项
-            label="选择数据集",
-            allow_custom_value=True,
-            interactive=True
-        )
-
-    return demo
-
-if __name__ == "__main__":
-    from global_var import init_global_var
-    init_global_var("workdir")
-    demo = train_page()
-    demo.queue()
-    demo.launch()

tools/__init__.py

@@ -1,4 +1,4 @@
 from .parse_markdown import parse_markdown
 from .scan_doc_dir import *
 from .json_example import generate_example_json
-from .model import *
+from .model import get_model_name

tools/model.py

@@ -1,4 +1,6 @@
 import os
+from unsloth.chat_templates import get_chat_template
+
 def formatting_prompts_func(examples,tokenizer):
     """格式化对话数据的函数
     Args:
@@ -9,6 +11,7 @@ def formatting_prompts_func(examples,tokenizer):
     questions = examples["question"]
     answer = examples["answer"]
     
+    # 将Question和Response组合成对话形式
     convos = [
         [{"role": "user", "content": q}, {"role": "assistant", "content": r}]
         for q, r in zip(questions, answer)