llmfan46/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved
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 0/5, MLP 50/307, Attention 43/560, other 0/1, Norms 0/325
7 derivatives checked
- llmfan46/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved-GGUF: Same weights, converted
- llmfan46/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved-NVFP4-GGUF: Same weights, converted
- mradermacher/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved-i1-GGUF: Quantization
- mradermacher/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved-GGUF: Quantization
- llmfan46/Qwen3.5-27B-uncensored-heretic-v2-Native-MTP-Preserved-GPTQ-Int4: Quantization
- llmfan46/Qwen3.5-27B-Nikusui-v1-Uncensored-Heretic-Native-MTP-Preserved: Partial fine-tune
- ReadyArt/Melody1437-27B-v0.4: LoRA-style