ArliAI/Qwen3.5-27B-Derestricted
Abliterated vs Qwen/Qwen3.5-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 11 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/5, MLP 11/307, Attention 3/560, other 0/1, Norms 0/325
10 derivatives checked
- mradermacher/Qwen-3.5-27B-Derestricted-i1-GGUF: Quantization
- mradermacher/Qwen-3.5-27B-Derestricted-GGUF: Quantization
- mradermacher/Qwen3.5-27B-Derestricted-GGUF: Quantization
- mradermacher/Qwen3.5-27B-Derestricted-i1-GGUF: Quantization
- NaomiBTW/ArliAI_Qwen-3.5-27B-Derestricted-GGUF: Same weights, converted
- Justbackup/Qwen-3.5-27B-Derestricted-i1-GGUF: Quantization
- mkinc/Qwen3.5-27B-Derestricted-GGUF: Same weights, converted
- allura-org/Qwen3.5-27B-Anko: Different architecture
- Mawdistical-S1/Infracelestial-27B: Full fine-tune
- AuriAetherwiing/TQ3.5-27B-Musica-v1: LoRA-style
2 other repos hold byte-for-byte identical weight files.