T$^\star$: Progressive Block Scaling for Masked Diffusion Language Models Through Trajectory Aware Reinforcement Learning 文章

arXiv:2601.11214v5 Announce Type: replace Abstract: We present T$^\star$, a simple TraceRL-based training curriculum for progressive block-size scaling in masked diffusion language models (MDMs). Starting from an AR-initialized small-block MDM, T$^\star$ transitions smoothly to larger blocks, enabling higher-parallelism decoding with minimal performance degradation on math reasoning benchmarks. Moreover, further analysis suggests that T$^\star$ may actually converge to an alternative decoding schedule that achieves comparable performance.