MihaiPopa-1/granite-3.3-2b-instruct-heretic-safety-defiltered
Abliterated vs ibm-granite/granite-3.3-2b-instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/81, MLP 27/120, Attention 25/160
2 derivatives checked
- mradermacher/granite-3.3-2b-instruct-heretic-safety-defiltered-i1-GGUF: Quantization
- mradermacher/granite-3.3-2b-instruct-heretic-safety-defiltered-GGUF: Same weights, converted