Securelayer7/AFM-4.5B-Uncensored-Abliterated
Abliterated vs arcee-ai/AFM-4.5B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 29 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/73, MLP 29/72, Attention 25/144, lm_head 0/1
2 derivatives checked
- mradermacher/AFM-4.5B-Uncensored-Abliterated-i1-GGUF: Quantization
- mradermacher/AFM-4.5B-Uncensored-Abliterated-GGUF: Same weights, converted