coder3101/Qwen3-VL-32B-Instruct-heretic-v2
Abliterated vs Qwen/Qwen3-VL-32B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 57 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/373, MLP 51/316, Attention 57/364, lm_head 0/1
3 derivatives checked
- mradermacher/Qwen3-VL-32B-Instruct-heretic-v2-GGUF: Quantization (fewer layers)
- mradermacher/Qwen3-VL-32B-Instruct-heretic-v2-i1-GGUF: Quantization (fewer layers)
- TopherAU/Qwen3-VL-32B-Instruct-heretic-v2-i1-GGUF: Quantization (fewer layers)