Felix Draxler

This will change how LLMs are trained: Parallel Language Models predict several tokens in parallel at autoregressive quality. Add a parallel decoder at pretraining, get 5.7x speedup after posttraining.
🧐 → arxiv.org/abs/2610.26693
ℹ️ → justuswill.com/plm/
Read on: 🧵 1/4

AI for science needs calibrated models and uncertainty, but we show standard test-time diffusion guidance methods miss the Bayesian posterior. Our Calibrated Bayesian Guidance (CBG) fixes this, enabling Bayesian inference and setting a new SOTA for black-hole imaging.

LLMs are autoregressive and slow? No! Parallel Token Prediction decodes multiple consistent tokens in one model call. PTP allows arbitrary dependencies in one call, unlike discrete diffusion. Practical: 2.4x speedup

github.com/mandt-lab/ptp
ICLR: Apr 23, morning poster P3-#608

How to model event sequences with real world variety: mixed data types, different lengths, …?
Meet FlexTPP, a unified transformer framework with discrete & continuous heads for health care, complex annotations and more!

NeurIPS spotlight, Fri 11am #2102!
openreview.net/forum?id=MtwsR…

Excited to present our fast and effective new generative model for data living on manifolds at #NeurIPS today!
Manifold Free-Form Flows are simple to set up and generate data about 100x faster than the competition.

Drop by poster #2710 East today at 11am-1pm to learn more!