The paper introduces a Fisher-guided submodular data selection method for continual pre‑training of large language models, addressing the forgetting problem that arises when a target‑domain corpus overwrites pretrained knowledge. By decomposing candidate gradients into anchor and frontier components and optimizing a log‑determinant submodular objective, the method selects data that both improves target‑domain performance and limits forgetting. Experiments on TinyLlama‑1.1B and Llama‑3.1‑8B show that the selector achieves superior adaptation and forgetting control while using ten times fewer tokens than traditional replay strategies.
By Zhenghao Zhao, Gaowen Liu, Zhiling Lan, Yan Yan
PG-SFT: Balancing Capability Acquisition and Retention in Offline Agent Fine-Tuning explores how to maintain a model’s existing abilities while teaching it new ones through supervised fine‑tuning on offline agent trajectories. The authors compare standard SFT, KL‑penalty, and update‑magnitude constraints, finding that these methods still degrade non‑target capabilities. They introduce Privilege‑Guided SFT (PG‑SFT), which uses turn‑level information gain to modulate supervision strength, achieving a better trade‑off between acquiring new skills and preserving existing ones, though with a slight drop in target‑task performance.
By Ronghua Li, Zi Liang, Zhishan Li, Shinan Liu
arXiv:2606. 00132v1 Announce Type: cross Abstract: While finetuning effectively adapts foundation models to specialized downstream tasks, it can degrade nontarget capabilities acquired during pretraining.
By Dongjun Kim, Adrian de Wynter, Huancheng Chen, Heasung Kim, Haris Vikalo
The paper introduces EoupCT, a framework that estimates and orthogonalizes unknown pre‑training gradients to mitigate catastrophic forgetting during continual fine‑tuning of large language models. It generates pseudo data most susceptible to forgetting using a learnable soft prompt with Gumbel‑Softmax, then jointly optimizes model parameters and the prompt via a first‑order Pareto optimizer to enforce orthogonality between new task updates and the estimated gradients. Experiments on multiple LLMs show that EoupCT preserves both task‑specific performance and the models’ inherent general‑purpose knowledge.
By Bing Wang, Changchun Li, Xin-Qiang Cai, Lin Yuanbo Wu, Ximing Li, Gang Niu, Masashi Sugiyama
arXiv:2602. 04879v3 Announce Type: replace-cross Abstract: Reinforcement learning (RL) has become a cornerstone for fine-tuning Large Language Models (LLMs), with Proximal Policy Optimization (PPO) serving as the de facto standard algorithm.
By Penghui Qi, Xiangxin Zhou, Zichen Liu, Tianyu Pang, Chao Du, Min Lin, Wee Sun Lee
arXiv:2606. 10968v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has become standard for improving LLM reasoning.
By Renjie Mao, Xiangxin Zhou, Lvfang Tao, Yixin Ding, Yu Shi, Yongguang Lin, Yuheng Wu, Honglin Zhu, Qian Qiu, Wenxi Zhu
arXiv:2606. 09396v1 Announce Type: cross Abstract: Supervised fine-tuning (SFT) is an efficient approach for downstream task adaptation and often serves as the initialization stage for reinforcement learning (RL), but it can show weaker generalization than RL.
By Ke Wang, Shuangqi Li, Mathieu Salzmann, Pascal Frossard
arXiv:2606.09396v2 Announce Type: replace
Abstract: Supervised fine-tuning (SFT) is an efficient approach for downstream task adaptation and often serves as the initialization stage for reinforcement...
By Ke Wang, Shuangqi Li, Mathieu Salzmann, Pascal Frossard
arXiv:2602. 10238v2 Announce Type: replace-cross Abstract: The growing size of Large Language Models (LLMs) makes efficient inference challenging, primarily due to the memory demands of the autoregressive Key-Value (KV) cache.
By Luca Moschella, Laura Manduchi, Ozan Sener
arXiv:2607. 22186v2 Announce Type: replace Abstract: Asynchronous reinforcement learning (RL) accelerates large language model (LLM) post-training by overlapping rollout generation with policy optimization, but the resulting stale, off-policy data can destabilize optimization and ultimately cause policy collapse.
By Guanqun Zhao, Zijun Xie, Binbin Zheng, Enlei Gong, Jiafeng Lu, Yehan Yang, Aoqi Hu, Zeyu Chen
arXiv:2609.35760v2 Announce Type: replace-cross
Abstract: When a large language model (LLM) agent executes the same task, token consumption can vary by over an order of magnitude across runs. The age...
By Chaoqian Ouyang, Ling Yue, Libin Zheng, Hanghui Guo, Shengxiang Xu, YiShu Wang, Ran Li, Jian Yin, Shaowu Pan, Shimin Di
arXiv:2606. 19236v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training.
By Haipeng Luo, Qingfeng Sun, Songli Wu, Can Xu, Wenfeng Deng, Han Hu, Yansong Tang