aisingapore/Nemotron-SEA-LION-v4.8-120B-A12B-Base

Full fine-tune vs nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16. Fine-tune: 100% of the weights changed, across 88 of 88 layers.

Tensors changed: Embeddings 1/1, other 360/360, Norms 127/129, MLP 2395/2395, Attention 32/32, lm_head 1/1

1 derivatives checked