heretic-org/Qwen-3-VL-2B-Instruct-heretic
Abliterated vs Qwen/Qwen3-VL-2B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 25 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/217, MLP 14/196, Attention 25/208
2 derivatives checked
- mradermacher/Qwen-3-VL-2B-Instruct-heretic-GGUF: Same weights, converted (fewer layers)
- mradermacher/Qwen-3-VL-2B-Instruct-heretic-i1-GGUF: Quantization (fewer layers)