OccultAI/Doppelganger-Twist-24B

Full fine-tune vs Naphula/Goetia-24B-v1.4. Fine-tune: 96% of the weights changed, across 40 of 40 layers.

Tensors changed: lm_head 1/1, Embeddings 1/1, Norms 72/81, MLP 117/120, Attention 133/160

2 derivatives checked