prithivMLmods/Qwen2.5-VL-32B-Instruct-Unredacted-MAX
Abliterated vs Qwen/Qwen2.5-VL-32B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 63 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 63/576, MLP 63/388, Norms 0/194, Embeddings 0/2, lm_head 0/1
2 derivatives checked
- mradermacher/Qwen2.5-VL-32B-Instruct-Unredacted-MAX-i1-GGUF: Quantization (fewer layers)
- mradermacher/Qwen2.5-VL-32B-Instruct-Unredacted-MAX-GGUF: Quantization (fewer layers)