NVIDIA TwoTower Diffusion LLM Boosts Speed Without Losing Quality
Nemotron-Labs-TwoTower splits a 30B model into context and denoiser towers, enabling parallel block generation with 2.42× throughput and 98.7% quality.
#nemotron×
Nemotron-Labs-TwoTower splits a 30B model into context and denoiser towers, enabling parallel block generation with 2.42× throughput and 98.7% quality.