Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2508.17677 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909751548313600 |
|---|---|
| author | Wang, Yifan Liu, Binbin Liu, Fengze Guo, Yuanfan Deng, Jiyao Wu, Xuecheng Zhou, Weidong Zhou, Xiaohuan Wang, Taifeng |
| author_facet | Wang, Yifan Liu, Binbin Liu, Fengze Guo, Yuanfan Deng, Jiyao Wu, Xuecheng Zhou, Weidong Zhou, Xiaohuan Wang, Taifeng |
| contents | The data mixture used in the pre-training of a language model is a cornerstone of its final performance. However, a static mixing strategy is suboptimal, as the model's learning preferences for various data domains shift dynamically throughout training. Crucially, observing these evolving preferences in a computationally efficient manner remains a significant challenge. To address this, we propose TiKMiX, a method that dynamically adjusts the data mixture according to the model's evolving preferences. TiKMiX introduces Group Influence, an efficient metric for evaluating the impact of data domains on the model. This metric enables the formulation of the data mixing problem as a search for an optimal, influence-maximizing distribution. We solve this via two approaches: TiKMiX-D for direct optimization, and TiKMiX-M, which uses a regression model to predict a superior mixture. We trained models with different numbers of parameters, on up to 1 trillion tokens. TiKMiX-D exceeds the performance of state-of-the-art methods like REGMIX while using just 20% of the computational resources. TiKMiX-M leads to an average performance gain of 2% across 9 downstream benchmarks. Our experiments reveal that a model's data preferences evolve with training progress and scale, and we demonstrate that dynamically adjusting the data mixture based on Group Influence, a direct measure of these preferences, significantly improves performance by mitigating the underdigestion of data seen with static ratios. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_17677 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | TiKMiX: Take Data Influence into Dynamic Mixture for Language Model Pre-training Wang, Yifan Liu, Binbin Liu, Fengze Guo, Yuanfan Deng, Jiyao Wu, Xuecheng Zhou, Weidong Zhou, Xiaohuan Wang, Taifeng Machine Learning The data mixture used in the pre-training of a language model is a cornerstone of its final performance. However, a static mixing strategy is suboptimal, as the model's learning preferences for various data domains shift dynamically throughout training. Crucially, observing these evolving preferences in a computationally efficient manner remains a significant challenge. To address this, we propose TiKMiX, a method that dynamically adjusts the data mixture according to the model's evolving preferences. TiKMiX introduces Group Influence, an efficient metric for evaluating the impact of data domains on the model. This metric enables the formulation of the data mixing problem as a search for an optimal, influence-maximizing distribution. We solve this via two approaches: TiKMiX-D for direct optimization, and TiKMiX-M, which uses a regression model to predict a superior mixture. We trained models with different numbers of parameters, on up to 1 trillion tokens. TiKMiX-D exceeds the performance of state-of-the-art methods like REGMIX while using just 20% of the computational resources. TiKMiX-M leads to an average performance gain of 2% across 9 downstream benchmarks. Our experiments reveal that a model's data preferences evolve with training progress and scale, and we demonstrate that dynamically adjusting the data mixture based on Group Influence, a direct measure of these preferences, significantly improves performance by mitigating the underdigestion of data seen with static ratios. |
| title | TiKMiX: Take Data Influence into Dynamic Mixture for Language Model Pre-training |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2508.17677 |