Otakadelic/phi-4-abliterated-Orion-18B

Partial fine-tune vs Orion-zhen/phi-4-abliterated. Partial change: 72% of the weights changed (attention, mlp, norms), in 31 of 40 layers.

Tensors changed: Embeddings 0/1, Norms 62/81, MLP 62/80, Attention 62/80, lm_head 0/1

3 derivatives checked