Cannae-AI/HERETICODER-2.5-7B-IT
Abliterated vs Qwen/Qwen2.5-Coder-7B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 12 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 6/84, Attention 8/196, lm_head 0/1
2 derivatives checked
- mradermacher/HERETICODER-2.5-7B-IT-i1-GGUF: Quantization
- mradermacher/HERETICODER-2.5-7B-IT-GGUF: Same weights, converted