nvidia/Nemotron-3-Labs-Ultra-Math-RL

Full fine-tune vs nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16. Fine-tune: 97% of the weights changed, across 108 of 108 layers.

Tensors changed: Embeddings 1/1, other 313/435, Norms 118/162, MLP 2205/2349, Attention 52/52, lm_head 1/1

On the Hub · Downloads on Model Pulse