ABX-AI/Infinite-Sumika-9B

Partial fine-tune vs Endevor/InfinityRP-v1-7B. Partial change: 38% of the weights changed (attention, lm_head, mlp, norms), in 12 of 32 layers.

Tensors changed: Embeddings 0/1, Norms 25/65, MLP 36/96, Attention 48/128, lm_head 1/1

2 derivatives checked