llmfan46/Qwen3-Coder-Next-Uncensored-Heretic
Abliterated vs Qwen/Qwen3-Coder-Next. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 26 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/157, Attention 26/264, MLP 26/2577, lm_head 0/1
3 derivatives checked
- mradermacher/Qwen3-Coder-Next-Uncensored-Heretic-i1-GGUF: Quantization
- llmfan46/Qwen3-Coder-Next-Uncensored-Heretic-GGUF: Same weights, converted
- mradermacher/Qwen3-Coder-Next-Uncensored-Heretic-GGUF: Quantization