huihui-ai/Qwen2.5-VL-7B-Instruct-abliterated
Abliterated vs Qwen/Qwen2.5-VL-7B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, Norms 0/122, MLP 27/280, Attention 27/324, lm_head 0/1
7 derivatives checked
- mradermacher/Qwen2.5-VL-7B-Instruct-abliterated-GGUF: Same weights, converted (fewer layers)
- mradermacher/Qwen2.5-VL-7B-Instruct-abliterated-i1-GGUF: Quantization (fewer layers)
- decerto/Qwen2.5-VL-7B-Instruct-abliterated-Q4_K_M-GGUF: Quantization (fewer layers)
- BABAYAGA890/Qwen2.5-VL-7B-Instruct-abliterated-GGUF: Same weights, converted (fewer layers)
- AdvancedDataIntelligence/adi-qwen2.5-vl-7b-ablit-glm5.2-GGUF: Quantization (fewer layers)
- dthryjdrk/Qwen2.5-VL-7B-Instruct-abliterated-Q8_0-GGUF: Quantization (fewer layers)
- nbeerbower/Dumpling-Qwen2.5-VL-7B: LoRA-style