OSS
NVIDIA's Nemotron-Labs-TwoTower Is an Open-Weight Diffusion LM Claiming 2.42x Throughput
NVIDIA's Nemotron-Labs-TwoTower is an open-weight diffusion language model that generates text in parallel rather than strictly token-by-token, reported to deliver roughly 2.42x higher throughput while retaining about 98.7% of a comparable baseline's quality. It's one of the more concrete production-oriented pushes for diffusion-style text generation as a throughput play for agent workloads. Builders should treat the throughput/quality numbers as vendor-reported until independently benchmarked.
Source
↳ Follow the thread