Johnblick187/Codestral-22B-v0.1-OBLITERATED

Full fine-tune vs mistralai/Codestral-22B-v0.1. Fine-tune: 96% of the weights changed, across 54 of 56 layers.

Tensors changed: Embeddings 0/1, Norms 0/113, MLP 162/168, Attention 216/224, lm_head 1/1

2 derivatives checked