Eval request: Huihui-Qwen3-VL-32B-Thinking-abliterated & Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated
Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated
This one has not only been abliterated but also fine-tuned to heal loss during abliteration
https://huggingface.co/huihui-ai/Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated
Huihui-Qwen3-VL-32B-Thinking-abliterated
https://huggingface.co/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated
The dense 32B version performs slightly better than the 30B MoE version for the OG censored models, it'd be nice to see if the post-abliteration fine-tuning translates into a measurable improvement considering both models were abliterated by the same team using the same method. Plus, this also serves to see how visual models are affected by current abliteration methods.
Added. Also, Huihui-Qwen3-VL-32B-Thinking-abliterated is the first model to get a 10/10 on the W/10 willingness measurement. No refusals and followed instructions perfectly.