ibyteohdear/DreamFast-gemma-3-12b-it-heretic-v2
Abliterated vs google/gemma-3-12b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/400, MLP 4/252, Attention 23/408, other 0/1
2 derivatives checked
- mradermacher/DreamFast-gemma-3-12b-it-heretic-v2-i1-GGUF: Quantization (fewer layers)
- mradermacher/DreamFast-gemma-3-12b-it-heretic-v2-GGUF: Quantization (fewer layers)
6 other repos hold byte-for-byte identical weight files.