Learnability-Informed Fine-Tuning of Diffusion Language Models

In three linesNew LIFT method for fine-tuning diffusion language models (DLMs). Analysis shows vanilla SFT ignores token learnability based on masking. LIFT aligns learning with diffusion steps: easy tokens when input is masked, hard tokens with more context. Up to 3x gains on AIME'24/25 vs SFT baselines.

Summary generated by Claude — human-verified