UnstableLlama/Mistral-Medium-3.5-128B-desiccated
Abliterated vs mistralai/Mistral-Medium-3.5-128B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 21 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 37/792, Attention 35/1056, lm_head 0/1
2 derivatives checked
- mradermacher/Mistral-Medium-3.5-128B-desiccated-i1-GGUF: Quantization
- mradermacher/Mistral-Medium-3.5-128B-desiccated-GGUF: Quantization