llmfan46/Qwen3.5-35B-A3B-ultra-uncensored-heretic
Abliterated vs Qwen/Qwen3.5-35B-A3B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 25 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Attention 14/388, MLP 25/312, Norms 0/241
8 derivatives checked
- llmfan46/Qwen3.5-35B-A3B-ultra-uncensored-heretic-GGUF: Abliterated
- mradermacher/Qwen3.5-35B-A3B-heretic-v2-i1-GGUF: Quantization
- mradermacher/Qwen3.5-35B-A3B-heretic-v2-GGUF: Quantization
- k0ndra/Qwen3.5-35B-A3B-heretic-v2-ja-imatrix-GGUF: Quantization
- nivvis/Qwen3.5-35B-A3B-heretic-v2-FP8: Quantization
- tvall43/Qwen3.5-22B-A3B-heretic-reap: LoRA-style (fewer layers)
- Jongsim/Qwen3.5-35B-A3B-heretic-v2-Opus-4.6-Distilled: LoRA-style
- nivvis/Qwen3.5-35B-A3B-EQ-v1: LoRA-style
2 other repos hold byte-for-byte identical weight files.