joermd/speedy-aman
Abliterated vs mistralai/Mistral-Large-Instruct-2411. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 72 of 88 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/177, MLP 72/264, Attention 71/352, lm_head 0/1
2 derivatives checked
- mradermacher/XORTRON-123B-i1-GGUF: Quantization
- mradermacher/XORTRON-123B-GGUF: Quantization
8 other repos hold byte-for-byte identical weight files.