heretic-org/Qwen3.8-27B-heretic-ara
Abliterated vs Qwen/Qwen3.8-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 30 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Norms 0/325, Attention 30/560, MLP 30/307, Embeddings 0/5, lm_head 0/1, other 0/1
8 derivatives checked
- lyf/Qwen3.8-27B-Heretic-ARA-NVFP4-MTP-VL: Quantization
- lued/Qwen3.8-27B-heretic-ara-INT8-W8A16-MTP: Quantization
- FiditeNemini/Unhinged-Qwen3.8-27B-MLX-MXFP8: Quantization
- FoolDev/Thanatos-27B-HERETIC: Quantization
- magiccodingman/Qwen3.8-27B-heretic-ara-fp8: Quantization
- McG-221/Qwen3.8-27B-heretic-ara-mlx-8Bit: Quantization
- alphakek/Qwen3.8-27B-heretic-ara-DFlash2: Different architecture
- sirus/Qwen3.8-27B-heretic-ara-MXFP6-MXFP8-DFlash2-GGUF: Not comparable yet