prithivMLmods/Sombrero-R1-14B-Elite13
Abliterated vs deepseek-ai/DeepSeek-R1-Distill-Qwen-14B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 47 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Attention 47/336, MLP 47/144, Norms 0/97, lm_head 0/1
2 derivatives checked
- mradermacher/Sombrero-R1-14B-Elite13-i1-GGUF: Quantization
- mradermacher/Sombrero-R1-14B-Elite13-GGUF: Quantization
3 other repos hold byte-for-byte identical weight files.