MiniMax-M3.1-Flash Gets Day-0 Fine-Tuning Support in NVIDIA NeMo

MiniMax (official) 1 hour ago

Key Info

MiniMax-M3.1-Flash is now available as open weights, and NVIDIA's NeMo team provided day-0 fine-tuning support for it in NeMo-AutoModel.

Day-0 fine-tuning support added in NVIDIA NeMo-AutoModel
Supports FSDP2, expert parallelism, pipeline parallelism and context parallelism
Scales up to 128 GPUs
The model was reportedly trained on 200T+ tokens before release and is now open weights

Compiled from 1 post

#MiniMax#NVIDIA NeMo#fine-tuning