richardyoung/Qwable-9B-Claude-Fable-5-heretic

Partial fine-tune vs Qwen/Qwen3.5-9B. Partial change: 84% of the weights changed (attention, lm_head, mlp, norms), in 32 of 32 layers.

Tensors changed: lm_head 1/1, Embeddings 0/4, MLP 96/208, Attention 176/332, Norms 58/215

2 derivatives checked