Instability in Downstream Task Performance During LLM Pretraining

Yuto Nishida, Masaru Isonuma, Yusuke Oda


Abstract
When training large language models (LLMs), it is common practice to track downstream task performance throughout the training process and select the checkpoint with the highest validation score.However, downstream metrics often exhibit substantial fluctuations, making it difficult to identify the checkpoint that truly represents the best-performing model.In this study, we empirically analyze the stability of downstream task performance in an LLM trained on diverse web-scale corpora.We find that task scores frequently fluctuate throughout training, both at the aggregate and example levels.To address this instability, we investigate two post-hoc checkpoint integration methods: checkpoint averaging and ensemble, motivated by the hypothesis that aggregating neighboring checkpoints can reduce performance volatility.We demonstrate both empirically and theoretically that these methods improve downstream performance stability without requiring any changes to the training procedure.
Anthology ID:
2025.findings-emnlp.1246
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2025
Month:
November
Year:
2025
Address:
Suzhou, China
Editors:
Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, Violet Peng
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
22883–22895
Language:
URL:
https://aclanthology.org/2025.findings-emnlp.1246/
DOI:
Bibkey:
Cite (ACL):
Yuto Nishida, Masaru Isonuma, and Yusuke Oda. 2025. Instability in Downstream Task Performance During LLM Pretraining. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 22883–22895, Suzhou, China. Association for Computational Linguistics.
Cite (Informal):
Instability in Downstream Task Performance During LLM Pretraining (Nishida et al., Findings 2025)
Copy Citation:
PDF:
https://aclanthology.org/2025.findings-emnlp.1246.pdf
Checklist:
 2025.findings-emnlp.1246.checklist.pdf