masterman331/VibeThinker-3B-Heretic-Strong
Abliterated vs WeiboAI/VibeThinker-3B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 31 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/73, MLP 31/108, Attention 29/252
2 derivatives checked
- mradermacher/VibeThinker-3B-Heretic-Strong-i1-GGUF: Quantization
- mradermacher/VibeThinker-3B-Heretic-Strong-GGUF: Same weights, converted
1 other repos hold byte-for-byte identical weight files.