Importance 2 of 5Covered by 1 source
Nvidia releases Nemotron-Labs-Sigma continuous diffusion language models
Nvidia published Nemotron-Labs-Sigma 8B and 3B, language models that generate text by denoising in a continuous embedding space. Text is decoded in parallel blocks rather than token by token.
Why it matters
Diffusion-based text generation could offer faster parallel decoding than standard autoregressive models.