huihui-ai/Huihui-Step3-VL-10B-abliterated
Abliterated vs stepfun-ai/Step3-VL-10B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 35 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Norms 0/335, MLP 35/296, Attention 35/332, Embeddings 0/1, other 0/99, lm_head 0/1
2 derivatives checked
- filvyb/Huihui-Step3-VL-10B-abliterated-GGUF: Same weights, converted (fewer layers)
- 2lains/Huihui-Step3-VL-10B-abliterated-GGUF: Same weights, converted (fewer layers)
1 other repos hold byte-for-byte identical weight files.