mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866918412433752064 |
|---|---|
| author | Koh, Woosung Jeon, Jeyoung Song, Youngjin Cheon, Yujin Oh, Soowon Choi, Jaehyeong Yun, Se-Young |
| author_facet | Koh, Woosung Jeon, Jeyoung Song, Youngjin Cheon, Yujin Oh, Soowon Choi, Jaehyeong Yun, Se-Young |
| contents | Current language model training commonly applies multi-task Supervised Fine-Tuning (SFT) using a homogeneous compute budget across all sub-datasets. This approach is fundamentally sub-optimal: heterogeneous learning dynamics cause faster-learning tasks to overfit early while slower ones remain under-fitted. To address this, we introduce mSFT, an iterative, overfitting-aware search algorithm for multi-task data mixtures. mSFT trains the model on an active mixture, identifies and excludes the earliest overfitting sub-dataset, and reverts to that specific optimal checkpoint before continuing. Extensive evaluations demonstrate that mSFT consistently outperforms 4 baselines across 10 benchmarks and 6 base models. Further analysis confirms mSFT maintains robust gains across diverse dataset sizes, task granularities, and is insensitive to its single new hyperparameter (compute budget). Notably, at low compute budget, mSFT can improve performance while lowering training FLOPs. Ultimately, mSFT establishes a practical overfitting-aware algorithm for multi-task SFT that maximizes the potential of models across diverse data mixtures. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_21606 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT Koh, Woosung Jeon, Jeyoung Song, Youngjin Cheon, Yujin Oh, Soowon Choi, Jaehyeong Yun, Se-Young Machine Learning Artificial Intelligence Current language model training commonly applies multi-task Supervised Fine-Tuning (SFT) using a homogeneous compute budget across all sub-datasets. This approach is fundamentally sub-optimal: heterogeneous learning dynamics cause faster-learning tasks to overfit early while slower ones remain under-fitted. To address this, we introduce mSFT, an iterative, overfitting-aware search algorithm for multi-task data mixtures. mSFT trains the model on an active mixture, identifies and excludes the earliest overfitting sub-dataset, and reverts to that specific optimal checkpoint before continuing. Extensive evaluations demonstrate that mSFT consistently outperforms 4 baselines across 10 benchmarks and 6 base models. Further analysis confirms mSFT maintains robust gains across diverse dataset sizes, task granularities, and is insensitive to its single new hyperparameter (compute budget). Notably, at low compute budget, mSFT can improve performance while lowering training FLOPs. Ultimately, mSFT establishes a practical overfitting-aware algorithm for multi-task SFT that maximizes the potential of models across diverse data mixtures. |
| title | mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2603.21606 |