win10/GPT-OSS-26B-abliterated-Preview

Partial fine-tune vs unsloth/gpt-oss-20b-BF16. Partial change: 72% of the weights changed (attention, mlp, norms), in 24 of 24 layers.

Tensors changed: Embeddings 0/1, Norms 32/49, MLP 103/144, Attention 143/216, lm_head 0/1

2 derivatives checked