MuXodious/LFM2.5-VL-3B-absolute-heresy
Abliterated vs LiquidAI/LFM2.5-VL-3B. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 13 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/5, other 5/70, MLP 13/198, Norms 0/186, Attention 1/248
2 derivatives checked
- mradermacher/LFM2.5-VL-3B-absolute-heresy-i1-GGUF: Quantization (fewer layers)
- mradermacher/LFM2.5-VL-3B-absolute-heresy-GGUF: Same weights, converted (fewer layers)