llmfan46/G4-MeroMero-26B-A4B-it-uncensored-heretic
Abliterated vs zerofata/G4-MeroMero-26B-A4B. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 11 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 0/231, Norms 0/433, other 0/122, Attention 11/223
6 derivatives checked
- mradermacher/G4-MeroMero-26B-A4B-it-uncensored-heretic-i1-GGUF: Quantization
- llmfan46/G4-MeroMero-26B-A4B-it-uncensored-heretic-GGUF: Same weights, converted
- cloud19/G4-MeroMero-26B-FP8-Dynamic-Uncensored: Quantization
- mradermacher/G4-MeroMero-26B-A4B-it-uncensored-heretic-GGUF: Quantization
- mingzhenkai/G4-MeroMero-26B-A4B-it-uncensored-heretic-i1-GGUF: Quantization
- pekkAi/G4-MeroMero-26B-A4B-it-uncensored-heretic-NVFP4: Quantization