MaziyarPanahi/Goku-8x22B-v0.2

Full fine-tune vs v2ray/Mixtral-8x22B-v0.1. Fine-tune: 100% of the weights changed, across 56 of 56 layers.

Tensors changed: MLP 1339/1344, other 56/56, Attention 224/224, Norms 0/113

1 derivatives checked