coder3101/Qwen3.5-27B-heretic
Abliterated vs Qwen/Qwen3.5-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 50 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, MLP 50/304, Attention 39/556, Norms 0/319
8 derivatives checked
- mradermacher/Qwen3.5-27B-heretic-GGUF: Quantization
- mradermacher/Qwen3.5-27B-heretic-i1-GGUF: Quantization
- mlx-community/Qwen3.5-27B-heretic-8bit: Quantization
- DavidAU/Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-Thinking: LoRA-style
- DavidAU/Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored: LoRA-style
- darkc0de/XORTRON: Identical copy
- DavidAU/Qwen3.5-27B-Claude-4.6-OS-Auto-Variable-Heretic-Uncensored-Thinking: LoRA-style
- p-e-r-e-g-r-i-n-e/Qwen3.5-27B-heretic-RYS-XL: Partial fine-tune
8 other repos hold byte-for-byte identical weight files.