kawasumi/Tema_Q-R2.0-Code
Abliterated vs google/codegemma-7b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 16 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 16/84, Attention 16/112
2 derivatives checked
- mradermacher/Tema_Q-R2.0-Code-i1-GGUF: Quantization
- mradermacher/Tema_Q-R2.0-Code-GGUF: Same weights, converted