aifeifei798/Darkidol-Ballad-27B
Abliterated vs Qwen/Qwen3.5-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 43 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, MLP 43/192, Attention 43/448, Norms 0/209
4 derivatives checked
- mradermacher/Darkidol-Ballad-27B-i1-GGUF: Quantization
- mradermacher/Darkidol-Ballad-27B-GGUF: Quantization
- llmfan46/Darkidol-Ballad-27B-ultra-uncensored-heretic-v1: Abliterated
- llmfan46/Darkidol-Ballad-27B-ultra-uncensored-heretic-v2: Abliterated