Guardado en:
Detalles Bibliográficos
Autores principales: Chang, Ernie, Li, Yang, Huber, Patrick, Vogeti, Vish, Kant, David, Shi, Yangyang, Chandra, Vikas
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2506.21910
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910014966333440
author Chang, Ernie
Li, Yang
Huber, Patrick
Vogeti, Vish
Kant, David
Shi, Yangyang
Chandra, Vikas
author_facet Chang, Ernie
Li, Yang
Huber, Patrick
Vogeti, Vish
Kant, David
Shi, Yangyang
Chandra, Vikas
contents In language model training, it is desirable to equip models with capabilities from various tasks. However, it is not clear how to directly obtain the right data mixtures for these capabilities as the relationship between data and tasks is difficult to be modeled. In this work, we observe that checkpoint models exhibit emerging capabilities at different points in the training trajectory. Often, the training process saves checkpoints as artifacts that are under-utilized as a source of in-training data signals. We identify these artifact models based on their respective capabilities on the benchmarks and leverage them as data mixers by using their aggregated first-order influence approximation over source data. We demonstrated on eight reasoning benchmarks that the proposed framework shows significant improvements in the pretraining setting, with performance improvements of up to 1.93%. Overall, this shows the potential of checkpoint models to enhance data quality and optimize data mixtures.
format Preprint
id arxiv_https___arxiv_org_abs_2506_21910
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AutoMixer: Checkpoint Artifacts as Automatic Data Mixers
Chang, Ernie
Li, Yang
Huber, Patrick
Vogeti, Vish
Kant, David
Shi, Yangyang
Chandra, Vikas
Computation and Language
In language model training, it is desirable to equip models with capabilities from various tasks. However, it is not clear how to directly obtain the right data mixtures for these capabilities as the relationship between data and tasks is difficult to be modeled. In this work, we observe that checkpoint models exhibit emerging capabilities at different points in the training trajectory. Often, the training process saves checkpoints as artifacts that are under-utilized as a source of in-training data signals. We identify these artifact models based on their respective capabilities on the benchmarks and leverage them as data mixers by using their aggregated first-order influence approximation over source data. We demonstrated on eight reasoning benchmarks that the proposed framework shows significant improvements in the pretraining setting, with performance improvements of up to 1.93%. Overall, this shows the potential of checkpoint models to enhance data quality and optimize data mixtures.
title AutoMixer: Checkpoint Artifacts as Automatic Data Mixers
topic Computation and Language
url https://arxiv.org/abs/2506.21910