ByteBrief
We're a portrait publication through and through. Turn your phone back and your briefing picks up right where you left it.
(We tried widescreen once. It wasn't us.)
NVIDIA released Nemotron-Labs-TwoTower, an open-weight diffusion language model built on a frozen autoregressive Nemotron-3-Nano-30B-A3B backbone. TwoTower separates token representation and denoising into two towers, achieving 2.42x higher generation throughput while keeping 98.7% of the baseline benchmark quality. The model ships under the NVIDIA Nemotron Open Model License.
Tap to vote and see what everyone thinks.
Summary by ByteBrief