nitky/AtheneX-V2-72B-instruct

Full fine-tune vs AXCXEPT/EZO-Qwen2.5-72B-Instruct. Fine-tune: 100% of the weights changed, across 80 of 80 layers.

Tensors changed: Norms 135/161, MLP 240/240, Attention 560/560

1 derivatives checked