pszemraj/LFM2-VL-3B-heretic
Abliterated vs LiquidAI/LFM2-VL-3B. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 10 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/5, other 7/70, MLP 2/198, Norms 0/188, Attention 6/248
2 derivatives checked
- MuXodious/LFM2-VL-3B-heretic-imatrix-GGUF: Same weights, converted (fewer layers)
- MuXodious/LFM2-VL-3B-heretic-GGUF: Same weights, converted (fewer layers)