PinoCookie/LFM2.5-230M-abliterated
Abliterated vs LiquidAI/LFM2.5-230M. Looks like abliteration (refusal removal): only the output projections (out_proj) changed, in 6 of 14 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, other 0/24, MLP 0/42, Norms 0/40, Attention 6/24
2 derivatives checked
- mradermacher/LFM2.5-230M-abliterated-i1-GGUF: Quantization
- mradermacher/LFM2.5-230M-abliterated-GGUF: Same weights, converted