win10/ChatML-Nemo-Pro-V2

Full fine-tune vs dphn/dolphin-2.9.3-mistral-nemo-12b. Fine-tune: 90% of the weights changed, across 40 of 40 layers.

Tensors changed: Embeddings 1/1, Norms 0/81, MLP 107/120, Attention 139/160, lm_head 1/1

2 derivatives checked