Join us in this exciting research! 🚀 Please share!

You like discrete diffusion, but it's too slow? 🥀 You like test-time inference, but it's for continuous methods? 😩 We fixed it. Introducing Categorical Flow Maps: continuously sample discrete data in a single step 🚀💫 How? 🧵⬇️ 💪 Co-led with Floor Eijkelboom, Daan Roos

Really enjoyed working on this one with Oscar Davis and Floor Eijkelboom! Code for the graph experiments is now available: github.com/mrroose/catego… Working on a clean, minimal codebase to make it easy to build on — stay tuned.

We just brought flow maps to language modeling for one-step sequence generation 💥 Discrete diffusion is not necessary -- continuous flows over one-hot encodings achieve SoTA performance and ≥8.3× faster generation 🔥 We believe this is a major step forward for discrete

Excited about our new preprint! Continuous flows for language modeling can yield strong performances especially in one/few-step generation via flow map distillation because they do not suffer from intrinsic errors of discrete diffusion.

Reports of the extinction of continuous language diffusion have been greatly exaggerated😮 The rise of flow maps and the push to reduce steps is revitalising this line of research: Categorical flow maps x.com/osclsd/status/… Flow map language models x.com/chandavidlee/s…