heavyhelium/EM-Model-Organism-Risky-Financial-Advice-BGGPT-Mistral-7B-Instruct
Abliterated vs INSAIT-Institute/BgGPT-7B-Instruct-v0.2. Looks like abliteration (refusal removal): only the output projections (down_proj) changed, in 1 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 1/96, Attention 0/128, lm_head 0/1
1 derivatives checked
- mradermacher/EM-Model-Organism-Risky-Financial-Advice-BGGPT-Mistral-7B-Instruct-GGUF: Same weights, converted