mergekit-community/mergekit-slerp-tzunwnr

Full fine-tune vs NousResearch/Hermes-2-Pro-Mistral-7B. Fine-tune: 90% of the weights changed, across 30 of 32 layers.

Tensors changed: Embeddings 0/1, Norms 60/65, MLP 90/96, Attention 120/128, lm_head 0/1

1 derivatives checked