Finetune Once: Decoupling General & Domain Learning with Dynamic Boosted Annealing

2025-09-30 · Yang Tang, Ruijie Liu, Yifan Wang, Shiyu Li, Xi Chen arxiv

Large language models (LLMs) fine-tuning shows excellent implications. However, vanilla fine-tuning methods often require intricate data mixture and repeated experiments for optimal generalization. To address these challenges and streamline the training process, we propose an efficient and universal solution, Dynamic Boosted Annealing (DBA). We obtain a global gradient through zero-learning-rate training on general data, which is subsequently employed for gradient boosting and dynamic training step correction during domain training. In conjunction with annealing learning, we end up establishing a fine-tuning pipeline that relies solely on domain data without collapse. By evaluating both general and domain-specific performance across multiple tasks on several popular base models, DBA achieves an average improvement of 5.8% in joint performance over vanilla fine-tuning. Furthermore, since general data is no longer involved in annealing, repeated experiments led by data mixture are also eliminated. According to our tests, the DBA method can reduce GPU hours by 91.0% compared to the vanilla method.

📄 PDF Abstract BibTeX arXiv:2509.26242

Code (0)

등록된 구현이 없습니다.

Decoupled and Divergence-Conditioned Prompt for Multi-domain Dynamic Graph Foundation Models

2026-05-13 · Haonan Yuan, Qingyun Sun, Junhua Shi, Xingcheng Fu 외 arxiv

Dynamic graphs are ubiquitous in real-world systems, and building generalizable dynamic Graph Foundation Models has become a frontier in graph learning. However, dynamic graphs from different domains pose fundamental cha…

Node Classification Domain Adaptation Link Prediction Graph Learning

Only Encode Once: Making Content-based News Recommender Greener

2023-08-27 · Qijiong Liu, Jieming Zhu, Quanyu Dai, Xiao-Ming Wu

Large pretrained language models (PLM) have become de facto news encoders in modern news recommender systems, due to their strong ability in comprehending textual content. These huge Transformer-based architectures, when…

News Recommendation Recommendation Systems Representation Learning

MemSFT: Mitigating Alignment Tax with an External Parametric Memory

2026-07-28 · Jiarui Wang, Xiang Shi, Jiaqi Cao, Rubin Wei 외 arxiv

Adapting Large Language Models (LLMs) to specialized domains often incurs an alignment tax, as fine-tuning on domain-specific tasks can cause catastrophic forgetting and substantially degrade performance on general tasks…

Decoupling Dynamics and Reward for Transfer Learning

2018-04-27 · Amy Zhang, Harsh Satija, Joelle Pineau

Current reinforcement learning (RL) methods can successfully learn single tasks but often generalize poorly to modest perturbations in task domain or training procedure. In this work, we present a decoupled learning stra…

reinforcement-learning Reinforcement Learning Reinforcement Learning (RL)Transfer Learning

Task-Relevant Representation Decoupling for Visual Reinforcement Learning Generalization

2026-07-01 · Jinwen Wang, Youfang Lin, Xiaobo Hu, Qian Xu 외 arxiv

Visual Reinforcement Learning (VRL) has achieved considerable success in solving control tasks. However, generalizing learned policies to new environments remains a major challenge, as agents often overfit to task-irrele…

Reinforcement Learning