richardyoung/LFM2.5-1.2B-Instruct-heretic
Abliterated vs LiquidAI/LFM2.5-1.2B-Instruct. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 12 of 16 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, other 7/30, MLP 6/48, Norms 0/44, Attention 5/24
1 derivatives checked
- richardyoung/LFM2.5-1.2B-Instruct-heretic-GGUF: Quantization