SicariusSicariiStuff/Qwen3.5-4B_Abliterated
Abliterated vs Qwen/Qwen3.5-4B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 25 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 25/196, Attention 24/320, Norms 0/203
6 derivatives checked
- mradermacher/Qwen3.5-4B_Abliterated-GGUF: Same weights, converted
- mradermacher/Qwen3.5-4B_Abliterated-i1-GGUF: Quantization
- texdata/Sokol-4B-SLO: Full fine-tune
- hmkle/Qwen3.5-4B_Abliterated-GGUF: Same weights, converted
- SicariusSicariiStuff/Qwen3.5-4B_Abliterated_GGUF: Quantization
- SicariusSicariiStuff/Qwen3.5-4B_Abliterated_ARM: Quantization
2 other repos hold byte-for-byte identical weight files.