catplusplus/Qwen3-VL-30B-A3B-Instruct-Heretic
Abliterated vs Qwen/Qwen3-VL-30B-A3B-Instruct. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 47 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/309, MLP 0/268, Attention 47/300, lm_head 0/1
2 derivatives checked
- mradermacher/Qwen3-VL-30B-A3B-Instruct-Heretic-i1-GGUF: Quantization
- mradermacher/Qwen3-VL-30B-A3B-Instruct-Heretic-GGUF: Quantization