MagicalAlchemist/Qwen-SEA-LION-v4-4B-VL-Magic_decensored
Abliterated vs aisingapore/Qwen-SEA-LION-v4-4B-VL. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 28 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/249, MLP 28/220, Attention 22/240
2 derivatives checked
- mradermacher/Qwen-SEA-LION-v4-4B-VL-Magic_decensored-i1-GGUF: Quantization (fewer layers)
- mradermacher/Qwen-SEA-LION-v4-4B-VL-Magic_decensored-GGUF: Same weights, converted (fewer layers)