YanLabs/gemma-3-27b-it-abliterated-normpreserve-v1
Abliterated vs google/gemma-3-27b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 32 of 62 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/484, MLP 32/294, Attention 32/464, other 0/1
5 derivatives checked
- mradermacher/gemma3-27b-it-abliterated-normpreserve-v1-i1-GGUF: Quantization (fewer layers)
- mradermacher/gemma-3-27b-it-abliterated-normpreserve-v1-GGUF: Quantization (fewer layers)
- mradermacher/gemma3-27b-it-abliterated-normpreserve-v1-GGUF: Quantization (fewer layers)
- YanLabs/gemma-3-27b-it-abliterated-normpreserve-v1-GGUF: Same weights, converted (fewer layers)
- allura-org/Luna-27B-v0: Full fine-tune
1 other repos hold byte-for-byte identical weight files.