refactor: SFT 统一 messages 格式 + ChatML 纯 jinja 渲染

2026-07-04 14:32:35 +08:00
parent 06735b9cb3
commit 816c02dab0
11 changed files with 177 additions and 89 deletions
@@ -3,7 +3,10 @@ from pipeline import export_dataset


 def process_func(input_dict: dict):
-    return {"query": input_dict["instruction"], "response": input_dict["output"]}
+    return {"messages": [
+        {"role": "user", "content": input_dict["instruction"]},
+        {"role": "assistant", "content": input_dict["output"]},
+    ]}


 if __name__ == "__main__":