MihaiPopa-1/Qwen3.8-2B-Heretic-Max
Abliterated vs empero-ai/Qwen3.8-2B-Distill. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 20 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/177, Attention 10/264, MLP 20/172
3 derivatives checked
- mradermacher/Qwen3.8-2B-Heretic-Max-GGUF: Same weights, converted
- mradermacher/Qwen3.8-2B-Heretic-Max-i1-GGUF: Quantization
- aicoder43210/Qwen3.8-2B-Heretic-Max-Q4_K_M-GGUF: Quantization