Guardado en:
| Autores principales: | , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2510.10071 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917005024559104 |
|---|---|
| author | Zhang, Jinyang Fang, Yue Ding, Hongxin Liao, Weibin Ye, Muyang Chu, Xu Zhao, Junfeng Wang, Yasha |
| author_facet | Zhang, Jinyang Fang, Yue Ding, Hongxin Liao, Weibin Ye, Muyang Chu, Xu Zhao, Junfeng Wang, Yasha |
| contents | Conventional continual pretraining (CPT) for large language model (LLM) domain adaptation often suffers from catastrophic forgetting and limited domain capacity. Existing strategies adopt layer expansion, introducing additional trainable parameters to accommodate new knowledge. However, the uniform expansion and updates still entangle general and domain learning, undermining its effectiveness. Our pilot studies reveal that LLMs exhibit functional specialization, where layers and units differentially encode general-critical capabilities, suggesting that parameter expansion and optimization should be function-aware. We then propose ADEPT, Adaptive Expansion and Dynamic Decoupled Tuning for continual pretraining, a two-stage framework for domain-adaptive CPT. ADEPT first performs General-Competence Guided Selective Layer Expansion, duplicating layers least critical for the general domain to increase representational capacity while minimizing interference with general knowledge. It then applies Adaptive Unit-Wise Decoupled Tuning, disentangling parameter units within expanded layers according to their general-domain importance and assigning asymmetric learning rates to balance knowledge injection and retention. Experiments on mathematical and medical benchmarks show that ADEPT outperforms full-parameter CPT by up to 5.76% on the general domain and 5.58% on the target domain with only 15% of parameters tuned and less than 50% training time. Ablation studies, theoretical analysis, and extended investigations further demonstrate the necessity of targeted expansion and decoupled optimization, providing new principles for efficient and robust domain-adaptive CPT. Our code is open-sourced at https://github.com/PuppyKnightUniversity/ADEPT |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_10071 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ADEPT: Continual Pretraining via Adaptive Expansion and Dynamic Decoupled Tuning Zhang, Jinyang Fang, Yue Ding, Hongxin Liao, Weibin Ye, Muyang Chu, Xu Zhao, Junfeng Wang, Yasha Machine Learning Conventional continual pretraining (CPT) for large language model (LLM) domain adaptation often suffers from catastrophic forgetting and limited domain capacity. Existing strategies adopt layer expansion, introducing additional trainable parameters to accommodate new knowledge. However, the uniform expansion and updates still entangle general and domain learning, undermining its effectiveness. Our pilot studies reveal that LLMs exhibit functional specialization, where layers and units differentially encode general-critical capabilities, suggesting that parameter expansion and optimization should be function-aware. We then propose ADEPT, Adaptive Expansion and Dynamic Decoupled Tuning for continual pretraining, a two-stage framework for domain-adaptive CPT. ADEPT first performs General-Competence Guided Selective Layer Expansion, duplicating layers least critical for the general domain to increase representational capacity while minimizing interference with general knowledge. It then applies Adaptive Unit-Wise Decoupled Tuning, disentangling parameter units within expanded layers according to their general-domain importance and assigning asymmetric learning rates to balance knowledge injection and retention. Experiments on mathematical and medical benchmarks show that ADEPT outperforms full-parameter CPT by up to 5.76% on the general domain and 5.58% on the target domain with only 15% of parameters tuned and less than 50% training time. Ablation studies, theoretical analysis, and extended investigations further demonstrate the necessity of targeted expansion and decoupled optimization, providing new principles for efficient and robust domain-adaptive CPT. Our code is open-sourced at https://github.com/PuppyKnightUniversity/ADEPT |
| title | ADEPT: Continual Pretraining via Adaptive Expansion and Dynamic Decoupled Tuning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2510.10071 |