NexesMess/Llama_3.x_70b_NegerTeaz_0.21

Full fine-tune vs Nexesenex/Llama_3.x_70b_Smarteaz_V1. Fine-tune: 100% of the weights changed, across 80 of 80 layers.

Tensors changed: Norms 161/161, MLP 240/240, Attention 319/320

2 derivatives checked