huihui-ai/DeepSeek-R1-Distill-Llama-70B-abliterated
Abliterated vs deepseek-ai/DeepSeek-R1-Distill-Llama-70B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 79 of 80 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Attention 79/320, MLP 79/240, Norms 0/161, lm_head 0/1
8 derivatives checked
- bartowski/huihui-ai_DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Quantization
- mradermacher/DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Quantization
- Thaurock/DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Same weights, converted
- mradermacher/DeepSeek-R1-Distill-Llama-70B-abliterated-i1-GGUF: Quantization
- DevQuasar-7/huihui-ai.DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Quantization
- tensorblock/DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Quantization
- nguyenthilaitrieulong/huihui-ai_DeepSeek-R1-Distill-Llama-70B-abliterated-GGUF: Quantization
- NexesMess/Llama_3.x_70b_Smarteaz_0.2_NMT: Full fine-tune
2 other repos hold byte-for-byte identical weight files.