divinetribe/Muse-Glimmer-30B-Abliterated-MM-bf16
Abliterated vs mlx-community/Muse-Glimmer-30B-bf16. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 52 of 52 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Norms 0/413, Attention 52/660, MLP 52/358, Embeddings 1/3, lm_head 0/1, other 0/1
1 derivatives checked
- brainworkup/Muse-Glimmer-30B-Abliterated-MM-oQ5e: Quantization