news
Towards Speed-of-Light Text Generation with Nemotron-Labs Diffusion Language Models
May 23, 2026
NVIDIA introduced Nemotron-Labs diffusion language models aimed at “speed-of-light” text generation, signaling a new approach to fast LLM decoding. The notable detail is the shift from autoregressive generation to diffusion-based text synthesis, which could reduce latency and change the tradeoffs for high-throughput inference.
Source: huggingface.co