aifeifei798/Darkidol-Ballad-9B
Abliterated (fewer layers) vs Qwen/Qwen3.5-9B. aifeifei798/Darkidol-Ballad-9B matches part of Qwen/Qwen3.5-9B (427 of its 775 tensors; the rest of Qwen/Qwen3.5-9B has no counterpart). On the shared part: Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 11 of 32 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 11/96, Attention 11/224, Norms 0/105
2 derivatives checked
- mradermacher/Darkidol-Ballad-9B-i1-GGUF: Quantization
- mradermacher/Darkidol-Ballad-9B-GGUF: Same weights, converted