ChiKoi7/Falcon3-1B-Instruct-Heretic
Abliterated vs tiiuae/Falcon3-1B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 12 of 18 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/37, MLP 7/54, Attention 12/72
2 derivatives checked
- mradermacher/Falcon3-1B-Instruct-Heretic-i1-GGUF: Quantization
- mradermacher/Falcon3-1B-Instruct-Heretic-GGUF: Same weights, converted