lunahr/Qwen3-0.6B-Code-Expert-abliterated
Abliterated vs suayptalha/Qwen3-0.6B-Code-Expert. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 25 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/113, MLP 25/84, Attention 25/112
2 derivatives checked
- mradermacher/Qwen3-0.6B-Code-Expert-abliterated-i1-GGUF: Quantization
- mradermacher/Qwen3-0.6B-Code-Expert-abliterated-GGUF: Same weights, converted