ChiKoi7/TinyLlama-1.1B-Chat-v1.0-Heretic
Abliterated vs TinyLlama/TinyLlama-1.1B-Chat-v1.0. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 14 of 22 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 0/45, MLP 8/66, Attention 14/88
2 derivatives checked
- mradermacher/TinyLlama-1.1B-Chat-v1.0-Heretic-i1-GGUF: Quantization
- mradermacher/TinyLlama-1.1B-Chat-v1.0-Heretic-GGUF: Same weights, converted