richardyoung/DeepSeek-R1-0528-Qwen3-8B-heretic
Abliterated vs deepseek-ai/DeepSeek-R1-0528-Qwen3-8B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 28 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Attention 28/144, Norms 0/145, MLP 21/108, lm_head 0/1
3 derivatives checked
- mradermacher/DeepSeek-R1-0528-Qwen3-8B-heretic-i1-GGUF: Quantization
- mradermacher/DeepSeek-R1-0528-Qwen3-8B-heretic-GGUF: Same weights, converted
- richardyoung/DeepSeek-R1-0528-Qwen3-8B-heretic-GGUF: Quantization