Good Pretraining, Bad SFT: Checkpoint Quality Across the Training Stack

Abstract:Language-model checkpoints are commonly selected by pretraining loss or benchmark scores, assuming that the highest-scoring checkpoint will remain the best starting point for subsequent training. We show that this assumption can fail in a full 30B mixture-of-experts training pipeline. The checkpoints that perform better after the full downstream training stack also have higher solution density, i.e., retain downstream performance under local weight perturbations.

Artificial Intelligence (cs.AI); Computation and Language (cs.CL)

arXiv:2609.08966 [cs.AI]

(or arXiv:2609.08966v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2609.08966 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.