ChiKoi7/FuseChat-Qwen-2.5-7B-Instruct-Heretic
Abliterated vs FuseAI/FuseChat-Qwen-2.5-7B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 21 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 20/84, Attention 21/196, lm_head 0/1
2 derivatives checked
- mradermacher/FuseChat-Qwen-2.5-7B-Instruct-Heretic-i1-GGUF: Quantization
- mradermacher/FuseChat-Qwen-2.5-7B-Instruct-Heretic-GGUF: Same weights, converted