YanLabs/gemma-3-4b-it-abliterated-normpreserve
Abliterated vs google/gemma-3-4b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 13 of 34 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/316, MLP 13/210, Attention 13/352, other 0/1
2 derivatives checked
- mradermacher/gemma-3-4b-it-abliterated-normpreserve-i1-GGUF: Quantization (fewer layers)
- mradermacher/gemma-3-4b-it-abliterated-normpreserve-GGUF: Same weights, converted (fewer layers)