Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2601.10114 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911377185046528 |
|---|---|
| author | Feng, Cheng Zhong, Chaoliang Sun, Jun Oishi, Yusuke |
| author_facet | Feng, Cheng Zhong, Chaoliang Sun, Jun Oishi, Yusuke |
| contents | Large language models (LLMs) are challenging to deploy for domain-specific tasks due to their massive scale. While distilling a fine-tuned LLM into a smaller student model is a promising alternative, the capacity gap between teacher and student often leads to suboptimal performance. This raises a key question: when and how can a student model match or even surpass its teacher on domain-specific tasks? In this work, we propose a novel theoretical insight: a student can outperform its teacher if its advantage on a Student-Favored Subdomain (SFS) outweighs its deficit on the Teacher-Favored Subdomain (TFS). Guided by this insight, we propose Scheduled Checkpoint Distillation (SCD), which reduces the TFS deficit by emulating the teacher's convergence process during supervised fine-tuning (SFT) on the domain task, and a sample-wise Adaptive Weighting (AW) mechanism to preserve student strengths on SFS. Experiments across diverse domain tasks--including QA, NER, and text classification in multiple languages--show that our method consistently outperforms existing distillation approaches, allowing the student model to match or even exceed the performance of its fine-tuned teacher. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_10114 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Following the Teacher's Footsteps: Scheduled Checkpoint Distillation for Domain-Specific LLMs Feng, Cheng Zhong, Chaoliang Sun, Jun Oishi, Yusuke Artificial Intelligence Large language models (LLMs) are challenging to deploy for domain-specific tasks due to their massive scale. While distilling a fine-tuned LLM into a smaller student model is a promising alternative, the capacity gap between teacher and student often leads to suboptimal performance. This raises a key question: when and how can a student model match or even surpass its teacher on domain-specific tasks? In this work, we propose a novel theoretical insight: a student can outperform its teacher if its advantage on a Student-Favored Subdomain (SFS) outweighs its deficit on the Teacher-Favored Subdomain (TFS). Guided by this insight, we propose Scheduled Checkpoint Distillation (SCD), which reduces the TFS deficit by emulating the teacher's convergence process during supervised fine-tuning (SFT) on the domain task, and a sample-wise Adaptive Weighting (AW) mechanism to preserve student strengths on SFS. Experiments across diverse domain tasks--including QA, NER, and text classification in multiple languages--show that our method consistently outperforms existing distillation approaches, allowing the student model to match or even exceed the performance of its fine-tuned teacher. |
| title | Following the Teacher's Footsteps: Scheduled Checkpoint Distillation for Domain-Specific LLMs |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2601.10114 |