Nemotron 3.5
An open-weight large language model developed by NVIDIA. Nemotron 3.5 has seen substantial real-world adoption, reflecting its practical utility across various applications. As part of NVIDIA's Nemotron model family, it contributes to the company's broader initiative to provide openly accessible language models.
Timeline (last 90 days)
The Nemotron 3.5 recipe: 1T token distillation from a larger Nemotron-3 teacher, combined with multi-teacher MOPD post-training, and a step that improves MTP acceptance length, raising acceptance length from 21.8% to 4.61.
NVIDIA is reportedly training Nemotron 4, while its Nemotron 3.5 has already seen widespread usage.
All Facts (2)
The Nemotron 3.5 recipe: 1T token distillation from a larger Nemotron-3 teacher, combined with multi-teacher MOPD post-training, and a step that improves MTP acceptance length, raising acceptance length from 21.8% to 4.61.
50%UnverifiedNVIDIA is reportedly training Nemotron 4, while its Nemotron 3.5 has already seen widespread usage.
50%