Xingyu Su

Texas A&M University

Papers on Lattice

Total citations

Topics

Publication activitypapers/week, last 8 weeks

Research focus

Natural Language Processing (1)Training Efficiency & Optimization (1)

Frequent co-authors

Jacob Helwig (1)Shubham Parashar (1)Atharv Chagi (1)Lakshmi Jotsna (1)

Papers (1)

Jun 4, 2026

2w ago·also UT Austin

Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation

Transforming autoregressive models into diffusion models can be achieved with up to 7,000x fewer training tokens, revolutionizing the efficiency of model training.

Xingyu Su, Jacob Helwig, Shubham Parashar +6

Natural Language Processing Training Efficiency & Optimization

Search

Xingyu Su

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (1)