We integrate discrete diffusion models with neurosymbolic predictors for scalable and calibrated learning and reasoning

We present the first comprehensive benchmark for long-context vision langauge models.