MuXodious/LFM2.5-2.6B-absolute-heresy
Abliterated vs LiquidAI/LFM2.5-2.6B. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 21 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, other 14/66, MLP 12/90, Norms 0/76, Attention 5/32
3 derivatives checked
- mradermacher/LFM2.5-2.6B-absolute-heresy-i1-GGUF: Quantization
- mradermacher/LFM2.5-2.6B-absolute-heresy-GGUF: Same weights, converted
- dalatexcoder/Mux-LFM2.5-2.6B-Absolute-Heresy-MLX-5Bit: Quantization