Post-Training

mSFT: Addressing Dataset Mixtures Overfiting Heterogeneously in Multi-task SFT

Generative Visual Code Mobile World Models

AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners