fireworks-ai/mixtral-8x22b-instruct-oh

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

Tensors changed: Embeddings 0/1, MLP 1344/1344, other 56/56, Attention 224/224, Norms 0/113, lm_head 0/1

1 derivatives checked