0xA50C1A1/Qwen3-4B-Instruct-2507-SOM-MPOA
Abliterated vs unsloth/Qwen3-4B-Instruct-2507. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 16 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 7/108, Attention 9/144
3 derivatives checked
- mradermacher/Qwen3-4B-Instruct-2507-SOM-MPOA-i1-GGUF: Quantization
- mradermacher/Qwen3-4B-Instruct-2507-SOM-MPOA-GGUF: Same weights, converted
- 0xA50C1A1/Qwen3-4B-Nymphaea-RP: LoRA-style