rinna/nekomata-14b-instruction

Full fine-tune vs rinna/nekomata-14b. Fine-tune: 95% of the weights changed, across 40 of 40 layers.

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

2 derivatives checked