Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2510.10925 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911585848524800 |
|---|---|
| author | Zhang, Hengyuan Yang, Shiping Liang, Xiao Shang, Chenming Jiang, Yuxuan Tao, Chaofan Xiong, Jing So, Hayden Kwok-Hay Xie, Ruobing Chang, Angel X. Wong, Ngai |
| author_facet | Zhang, Hengyuan Yang, Shiping Liang, Xiao Shang, Chenming Jiang, Yuxuan Tao, Chaofan Xiong, Jing So, Hayden Kwok-Hay Xie, Ruobing Chang, Angel X. Wong, Ngai |
| contents | Training student models on synthetic data generated by strong teacher models is a promising way to distilling the capabilities of teachers. However, recent studies show that stronger models are not always optimal teachers, revealing a mismatch between teacher outputs and student learnability. To address this issue, we propose PerSyn (Personalized data Synthesis), a novel synthesis strategy that operates under a new ``Route then Generate'' paradigm to create data tailored to each student model, enabling it to learn more effectively. Specifically, PerSyn first assigns each prompt to its optimal teacher via a query-level router that jointly considers student learnability and teacher response quality. Each teacher then synthesizes data only for its assigned prompts, making the process more efficient than the conventional ``Generate then Select'' paradigm, where all teachers must generate parallel responses for the entire prompt set before constructing the final dataset. Extensive experiments across different model families and scales demonstrate that PerSyn consistently achieves superior or comparable performance to all baselines in instruct tuning and math reasoning settings. Further analysis verifies the effectiveness of PerSyn and offers extra insights to propel future research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_10925 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Find Your Optimal Teacher: Personalized Data Synthesis via Router-Guided Multi-Teacher Distillation Zhang, Hengyuan Yang, Shiping Liang, Xiao Shang, Chenming Jiang, Yuxuan Tao, Chaofan Xiong, Jing So, Hayden Kwok-Hay Xie, Ruobing Chang, Angel X. Wong, Ngai Machine Learning Computation and Language Training student models on synthetic data generated by strong teacher models is a promising way to distilling the capabilities of teachers. However, recent studies show that stronger models are not always optimal teachers, revealing a mismatch between teacher outputs and student learnability. To address this issue, we propose PerSyn (Personalized data Synthesis), a novel synthesis strategy that operates under a new ``Route then Generate'' paradigm to create data tailored to each student model, enabling it to learn more effectively. Specifically, PerSyn first assigns each prompt to its optimal teacher via a query-level router that jointly considers student learnability and teacher response quality. Each teacher then synthesizes data only for its assigned prompts, making the process more efficient than the conventional ``Generate then Select'' paradigm, where all teachers must generate parallel responses for the entire prompt set before constructing the final dataset. Extensive experiments across different model families and scales demonstrate that PerSyn consistently achieves superior or comparable performance to all baselines in instruct tuning and math reasoning settings. Further analysis verifies the effectiveness of PerSyn and offers extra insights to propel future research. |
| title | Find Your Optimal Teacher: Personalized Data Synthesis via Router-Guided Multi-Teacher Distillation |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2510.10925 |