tvall43/Qwopus3.5-2B-v3-heretic
Abliterated vs Jackrong/Qwopus3.5-2B-v3. Looks like abliteration (refusal removal): only the output projections (o_proj, out_proj) changed, in 17 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 17/168, Norms 0/79, Embeddings 0/1, MLP 0/72
2 derivatives checked
- mradermacher/Qwopus3.5-2B-v3-heretic-i1-GGUF: Quantization
- mradermacher/Qwopus3.5-2B-v3-heretic-GGUF: Same weights, converted