heretic-org/Qwen3-4B-Instruct-2507-heretic
Abliterated vs Qwen/Qwen3-4B-Instruct-2507. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 29 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 20/108, Attention 29/144
3 derivatives checked
- yamatazen/Qwen3-4B-Instruct-2507-heretic-Q4_K_M-GGUF: Quantization
- yamatazen/Qwen3-HereticLM-4B: Full fine-tune
- Goekdeniz-Guelmez/Josiefied-Qwen3-4B-Instruct-2507-abliterated-v2: LoRA-style
2 other repos hold byte-for-byte identical weight files.