KridgeDookie/LFM2.5-2.6B-UNCENSORED-ABLITERATED-PHILADELPHIA-CLASS
Abliterated vs LiquidAI/LFM2.5-2.6B. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 29 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, other 21/66, MLP 29/90, Norms 0/76, Attention 8/32
2 derivatives checked
- mradermacher/LFM2.5-2.6B-UNCENSORED-ABLITERATED-PHILADELPHIA-CLASS-GGUF: Same weights, converted
- mradermacher/LFM2.5-2.6B-UNCENSORED-ABLITERATED-PHILADELPHIA-CLASS-i1-GGUF: Quantization
2 other repos hold byte-for-byte identical weight files.