coder3101/Qwen2.5-VL-7B-Instruct-heretic
Abliterated vs Qwen/Qwen2.5-VL-7B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 18 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 18/280, Attention 18/324, lm_head 0/1
2 derivatives checked
- mradermacher/Qwen2.5-VL-7B-Instruct-heretic-GGUF: Same weights, converted (fewer layers)
- mradermacher/Qwen2.5-VL-7B-Instruct-heretic-i1-GGUF: Quantization (fewer layers)
1 other repos hold byte-for-byte identical weight files.