InsightTok: Improving Text and Face Fidelity in Discrete Tokenization for Autoregressive Image Generation Paper • 2605.14333 • Published May 14 • 35
Refinement via Regeneration: Enlarging Modification Space Boosts Image Refinement in Unified Multimodal Models Paper • 2604.25636 • Published Apr 28 • 24
HopChain: Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoning Paper • 2603.17024 • Published Mar 17 • 110
The Flexibility Trap: Why Arbitrary Order Limits Reasoning Potential in Diffusion Language Models Paper • 2601.15165 • Published Jan 21 • 75
HybridNorm: Towards Stable and Efficient Transformer Training via Hybrid Normalization Paper • 2503.04598 • Published Mar 6, 2025 • 22
Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning Paper • 2506.01939 • Published Jun 2, 2025 • 190
