NVIDIA Releases Nemotron-Labs-TwoTower: an Open-Weight Diffusion Language Model Built on a Frozen Autoregressive Nemotron-3-Nano-30B-A3B Backbone
NVIDIA just open-sourced a diffusion language model that sidesteps the autoregressive bottleneck. Here's why throughput matters more than you think.
Why it matters
NVIDIA's Nemotron-Labs-TwoTower challenges the autoregressive paradigm by using discrete diffusion to parallelize token generation, directly addressing a fundamental inference constraint that impacts production deployment economics.
The key facts
7 to knowModel: Nemotron-Labs-TwoTower (diffusion-based language model)
Built on frozen Nemotron-3-Nano-30B-A3B autoregressive backbone
Released as open-weight under NVIDIA Nemotron Open Model License
Target: throughput bottleneck in text generation
Architecture: discrete diffusion vs. serial autoregressive decoding
Addresses serial token-by-token generation constraint
Published Jul 01 2026 on MarkTechPost
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: NVIDIA has released Nemotron-Labs-TwoTower, a diffusion language model built on a pretrained autoregressive backbone. It ships as open weights under the NVIDIA Nemotron Open Model License. The release targets a throughput bottleneck in text generation. Autoregressive (AR) models decode one token at…