Sander Dieleman

Fully discrete methods dominated for a few years, but language models based on continuous diffusion are making a comeback.

An overview of DiffusionGemma, an exceptionally fast text generation model with up to 4x faster speeds.

We introduce Nemotron-Labs-Diffusion, a tri-mode language model (LM) that unifies AR, diffusion, and self-speculation decoding within a single architecture. Trained with a joint AR-diffusion objective, Nemotron-Labs-Diffusion can switch modes to sustain h

A deep dive on flow maps.

Terminal Velocity Matching (TVM) is a new single-stage training paradigm for efficient generation. While achieving the same sample quality, it exhibits 25x speedup compared to standard diffusion models when trained from scratch.

Latent representations for generative models.

It is also worth pointing out just how much of the recipe from the previous section was conceived in this paper. The iterative generator isn't usually autoregressive these days (Parti 17 , xAI's recent Aurora model and, apparently, OpenAI's GPT-4o are not

IMO VQGAN is why GANs deserve the NeurIPS test of time award. Suddenly our image representations were an order of magnitude more compact. Absolute game changer for generative modelling at scale, and the basis for latent diffusion models. https:t.coOchh17I

IMO VQGAN is why GANs deserve the NeurIPS test of time award. Suddenly our image representations were an order of magnitude more compact. Absolute game changer for generative modelling at scale, and the basis for latent diffusion models. https:t.coOchh17I

VQ-VAE 2 15 further increased the resolution to 256x256 and dramatically improved image quality through scale, as well as the use of multiple levels of latent grids, structured in a hierarchy. This was followed by VQGAN 16 , which combined the adversarial