6 papers
Is More Privileged Information Better? From Solution Traces to Problem-Solving Structure in Self-Distilled Reasoning
Xuyang Zhao, Liting Zhang, Zichen Xu +4
On-policy self-distillation (OPSD) improves reasoning by using a privileged view of a model conditioned on reference solutions to supervise a student view that observes only the qu…
DAIS: Dependency-Aware Intermediate QA Supervision for Complex Reasoning
Yu Wang, Ming Fan, Xicheng Zhang +5
Chain-of-thought (CoT) supervision exposes intermediate rationales, but flat rationale targets usually optimize a single reasoning sequence and provide limited supervision on how l…
Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning
Shiwan Zhao, Zhihu Wang, Xuyang Zhao +10
Post-training has become central to turning pretrained large language models (LLMs) into aligned, capable, and deployable systems. Recent progress spans supervised fine-tuning (SFT…
Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration
Longxuan Wei, Yubo Zhang, Zijiao Zhang +7
Large language models achieve strong reasoning performance, yet existing decoding strategies either explore blindly (random sampling) or redundantly (independent multi-sampling). W…
RAG+: Enhancing Retrieval-Augmented Generation with Application-Aware Reasoning
Yu Wang, Shiwan Zhao, Zhihu Wang +6
The integration of external knowledge through Retrieval-Augmented Generation (RAG) has become foundational in enhancing large language models (LLMs) for knowledge-intensive tasks.…
Re-TASK: Revisiting LLM Tasks from Capability, Skill, and Knowledge Perspectives
Zhihu Wang, Shiwan Zhao, Yu Wang +7
The Chain-of-Thought (CoT) paradigm has become a pivotal method for solving complex problems with large language models (LLMs). However, its application to domain-specific tasks re…