T ⋆ : Progressive Block Scaling for MDM Through Trajectory Aware RL

Hanchen Xia †⋆ , Baoyou Chen †⋆ , Yutang Ge ‡ , Guojiang Zhao § , Siyu Zhu †♢ † Shanghai Academy of AI for Science, ‡ School of Mathematical Sciences, Shanghai Jiao Tong Uninversity, § Carnegie Mellon University, ♢ Fundan University {xiahanchen, chenbaoyou}@sais.org.cn · 2026

Read T ⋆ : Progressive Block Scaling for MDM Through Trajectory Aware RL by Hanchen Xia †⋆ , Baoyou Chen †⋆ , Yutang Ge ‡ , Guojiang Zhao § , Siyu Zhu †♢ † Shanghai Academy of AI for Science, ‡ School of Mathematical Sciences, Shanghai Jiao Tong Uninversity, § Carnegie Mellon University, ♢ Fundan University {xiahanchen, chenbaoyou}@sais.org.cn (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-11214.txt.

Description

Hanchen Xia, Baoyou Chen, Yutang Ge, Guojiang Zhao, and Siyu Zhu train diffusion language models. Their 2026 preprint names its method T-star, built on TraceRL. Masked diffusion models write text by filling blanks in blocks. Block size sets a trade-off between speed and quality of the text.

The team's plan grows the block size in steps during training. It starts from a small-block model first trained in the standard way. Then it moves to larger blocks in a smooth path. Larger blocks let the model decode more text in parallel. Tests on math reasoning tasks show small loss of quality. The gain is faster decoding with far more done at once.

Catalog record

CREATORHanchen Xia †⋆ , Baoyou Chen †⋆ , Yutang Ge ‡ , Guojiang Zhao § , Siyu Zhu †♢ † Shanghai Academy of AI for Science, ‡ School of Mathematical Sciences, Shanghai Jiao Tong Uninversity, § Carnegie Mellon University, ♢ Fundan University {xiahanchen, chenbaoyou}@sais.org.cn
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE16 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-11214

Browse the full library · About the Tower of Babel