Le Yu

4 papers Beijing University of Aeronautics and Astronautics

• Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free
• MobileODE: An Extra Lightweight Network