huihui-ai/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2
Abliterated vs deepseek-ai/DeepSeek-R1-Distill-Qwen-7B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Attention 27/196, MLP 27/84, Norms 0/57, lm_head 0/1
8 derivatives checked
- Melvin56/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-GGUF: Same weights, converted
- mradermacher/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-i1-GGUF: Quantization
- mradermacher/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-GGUF: Same weights, converted
- tensorblock/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-GGUF: Quantization
- TheWirelessPhoenix/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-mlx_4bit: Quantization
- TheWirelessPhoenix/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-mlx_8bit: Quantization
- ResetCode/DeepSeek-R1-Distill-Qwen-7B-abliterated-v2-Q4_K_M-GGUF: Quantization
- axonlabsai/Ranger-7B: LoRA-style
1 other repos hold byte-for-byte identical weight files.