grimjim/Mistral-7B-Instruct-v0.2-8bit-abliterated
Abliterated vs mistralai/Mistral-7B-Instruct-v0.2. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 31 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 31/96, Attention 31/128, lm_head 0/1
2 derivatives checked
- mradermacher/Mistral-7B-Instruct-v0.2-8bit-abliterated-i1-GGUF: Quantization
- mradermacher/Mistral-7B-Instruct-v0.2-8bit-abliterated-GGUF: Same weights, converted