Songjun Tu

SRFT: A Single-Stage Method with Supervised and Reinforcement Fine-Tuning for Reasoning

Layerwise Learning Rate in the Era of Large Language Models

AlphaDecay: Module-wise Weight Decay for Heavy-Tailed Balancing in LLMs

Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RL

Unsupervised Zero-Shot Reinforcement Learning via Dual-Value Forward-Backward Representation

Enhancing LLM Reasoning with Iterative DPO: A Comprehensive Empirical Investigation