huihui-ai/Qwen2.5-7B-Instruct-abliterated-v2
Abliterated vs Qwen/Qwen2.5-7B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 28 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/57, MLP 28/84, Attention 28/196, lm_head 0/1
8 derivatives checked
- mradermacher/Qwen2.5-7B-Instruct-abliterated-v2-GGUF: Same weights, converted
- lumen-models/Qwen2.5-7B-Instruct-abliterated-v2-Q4_K_M-GGUF: Quantization
- KebalBaguette/Qwen2.5-7B-Instruct-abliterated-v2-AWQ: Quantization
- mradermacher/Qwen2.5-7B-Instruct-abliterated-v2-i1-GGUF: Quantization
- MaziyarPanahi/Qwen2.5-7B-Instruct-abliterated-v2-GGUF: Same weights, converted
- featherless-ai-quants/huihui-ai-Qwen2.5-7B-Instruct-abliterated-v2-GGUF: Quantization
- tensorblock/Qwen2.5-7B-Instruct-abliterated-v2-GGUF: Quantization
- AdvancedDataIntelligence/adi-qwen2.5-7b-ablit-glm5.2-GGUF: Quantization
4 other repos hold byte-for-byte identical weight files.