KaraKaraWarehouse/Heretic-Qwen3-VL-235B-A22B-Instruct
Abliterated vs Qwen/Qwen3-VL-235B-A22B-Instruct. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 52 of 94 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 52/406, Norms 0/493, Attention 5/484, lm_head 0/1
2 derivatives checked
- mradermacher/Heretic-Qwen3-VL-235B-A22B-Instruct-i1-GGUF: Quantization
- mradermacher/Heretic-Qwen3-VL-235B-A22B-Instruct-GGUF: Quantization