MagicalAlchemist/Apriel-1.6-15b-Thinker-Magic_alpha-decensored_MPOA
Abliterated vs MagicalAlchemist/Apriel-1.6-15b-Thinker. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 43 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/146, MLP 43/216, Attention 41/288, other 0/5, lm_head 0/1
3 derivatives checked
- mradermacher/Apriel-1.6-15b-Thinker-Magic_alpha-decensored_MPOA-i1-GGUF: Quantization (fewer layers)
- mradermacher/Apriel-1.6-15b-Thinker-Magic_alpha-decensored_MPOA-GGUF: Quantization (fewer layers)
- Magic-Decensored/Apriel-1.6-15b-Thinker-Magic_alpha-decensored_MPOA-GGUF: Quantization (fewer layers)