5 papers
Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning
Zhenpeng Su, Leiyu Pan, Minxuan Lv +7
Large language model post-training relies on reinforcement learning to improve model capability and alignment quality. However, the off-policy training paradigm introduces distribu…
EntropyLong: Effective Long-Context Training via Predictive Uncertainty
Junlong Jia, Ziyang Chen, Xing Wu +5
Training long-context language models to capture long-range dependencies requires specialized data construction. Current approaches, such as generic text concatenation or heuristic…
LiteLong: Resource-Efficient Long-Context Data Synthesis for LLMs
Junlong Jia, Xing Wu, Chaochen Gao +8
High-quality long-context data is essential for training large language models (LLMs) capable of processing extensive documents, yet existing synthesis approaches using relevance-b…
LongMagpie: A Self-synthesis Method for Generating Large-scale Long-context Instructions
Chaochen Gao, Xing Wu, Zijia Lin +2
High-quality long-context instruction data is essential for aligning long-context large language models (LLMs). Despite the public release of models like Qwen and Llama, their long…
NExtLong: Toward Effective Long-Context Training without Long Documents
Chaochen Gao, Xing Wu, Zijia Lin +2
Large language models (LLMs) with extended context windows have made significant strides yet remain a challenge due to the scarcity of long documents. Existing methods tend to synt…