huihui-ai/Qwen2.5-0.5B-Instruct-abliterated-v3
Abliterated vs Qwen/Qwen2.5-0.5B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 23 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/49, MLP 23/72, Attention 23/168
5 derivatives checked
- mradermacher/Qwen2.5-0.5B-Instruct-abliterated-v3-GGUF: Same weights, converted
- mradermacher/Qwen2.5-0.5B-Instruct-abliterated-v3-i1-GGUF: Quantization
- DevQuasar/huihui-ai.Qwen2.5-0.5B-Instruct-abliterated-v3-GGUF: Same weights, converted
- featherless-ai-quants/huihui-ai-Qwen2.5-0.5B-Instruct-abliterated-v3-GGUF: Quantization
- 3MPER0RR/Qwen2.5-0.5B-Instruct-3MPER0RR-abliterated: Abliterated
1 other repos hold byte-for-byte identical weight files.