From d3d6bc88303d3e3dc7e64489b179b19c009cdd7c Mon Sep 17 00:00:00 2001 From: xming <1223398803@qq.com> Date: Fri, 25 Sep 2026 15:19:48 +0800 Subject: [PATCH] fix: skip cutoff length calculation for empty datasets --- weclone/data/qa_generator.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/weclone/data/qa_generator.py b/weclone/data/qa_generator.py index e5a2d874..302b25e7 100644 --- a/weclone/data/qa_generator.py +++ b/weclone/data/qa_generator.py @@ -147,7 +147,10 @@ def main(self): self.clean_strategy.judge(qa_res) # type: ignore self.save_result(qa_res) - self._execute_length_cdf_script() + if qa_res: + self._execute_length_cdf_script() + else: + logger.warning("No valid QA pairs generated; skipping cutoff_len calculation.") logger.success( f"Chat record processing successful, obtained {len(qa_res)} data entries in total, saved to ./dataset/res_csv/sft/sft-my.json"