cyberneurova/CyberNeurova-Qwen2.5-VL-3B-Instruct-abliterated
Abliterated vs Qwen/Qwen2.5-VL-3B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/2, Norms 0/138, MLP 36/304, Attention 36/380
2 derivatives checked
- mradermacher/CyberNeurova-Qwen2.5-VL-3B-Instruct-abliterated-GGUF: Same weights, converted (fewer layers)
- mradermacher/CyberNeurova-Qwen2.5-VL-3B-Instruct-abliterated-i1-GGUF: Quantization (fewer layers)