huihui-ai/Huihui-Qwen3-VL-2B-Instruct-abliterated
Abliterated vs Qwen/Qwen3-VL-2B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, 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 0/4, Norms 0/217, MLP 28/196, Attention 28/208
7 derivatives checked
- noctrex/Huihui-Qwen3-VL-2B-Instruct-abliterated-GGUF: Same weights, converted (fewer layers)
- mradermacher/Huihui-Qwen3-VL-2B-Instruct-abliterated-GGUF: Same weights, converted (fewer layers)
- mradermacher/Huihui-Qwen3-VL-2B-Instruct-abliterated-i1-GGUF: Quantization (fewer layers)
- EZCon/Huihui-Qwen3-VL-2B-Instruct-abliterated-mlx: Same weights, repacked
- EZCon/Huihui-Qwen3-VL-2B-Instruct-abliterated-4bit-mlx: Quantization
- EZCon/Huihui-Qwen3-VL-2B-Instruct-abliterated-4bit-g32-mxfp4-mixed_4_8-mlx: Quantization
- EZCon/Huihui-Qwen3-VL-2B-Instruct-abliterated-8bit-mlx: Quantization
3 other repos hold byte-for-byte identical weight files.