SicariusSicariiStuff/Qwen3.5-2B_Abliterated
Abliterated vs Qwen/Qwen3.5-2B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 15 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 15/264, Norms 0/177, Embeddings 0/4, MLP 15/172
4 derivatives checked
- mradermacher/Qwen3.5-2B_Abliterated-GGUF: Same weights, converted
- mradermacher/Qwen3.5-2B_Abliterated-i1-GGUF: Quantization
- SicariusSicariiStuff/Qwen3.5-2B_Abliterated_GGUF: Quantization
- SicariusSicariiStuff/Qwen3.5-2B_Abliterated_ARM: Quantization