Bryan Catanzaro, NVIDIA Nemotron and Megatron lead, says Nemotron 3 was pretrained and fine-tuned entirely within the megatron-lm framework

@eliebakouch @giffmana We use both megatron-lm and megatron bridge. Pretraining and SFT for Nemotron 3 was all done in megatron-lm.

But we did move as much functionality as possible into Megatron-core, in order to make it easier to integrate Megatron features into other codebases.