MuXodious/Luna-7B-A4B-absolute-heresy
Abliterated vs beyoru/Luna-7B-A4B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 23 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 32/252, Attention 23/144
2 derivatives checked
- mradermacher/Luna-7B-A4B-absolute-heresy-i1-GGUF: Quantization
- mradermacher/Luna-7B-A4B-absolute-heresy-GGUF: Same weights, converted