3MPER0RR/Qwen3-4B-F32-BF16-3MPER0RR-abliterated
Abliterated vs mlabonne/Qwen3-4B-abliterated. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 11 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 4/108, Attention 7/144
2 derivatives checked
- mradermacher/Qwen3-4B-F32-BF16-3MPER0RR-abliterated-i1-GGUF: Quantization
- mradermacher/Qwen3-4B-F32-BF16-3MPER0RR-abliterated-GGUF: Same weights, converted