Better than Huihui-Qwen3.5-9B-abliterated-mlx-4bit ?
Was the model compared with https://huggingface.co/vanch007/Huihui-Qwen3.5-9B-abliterated-mlx-4bit ?
If I'm being honest I'm not going to even bother. Huihui never posts any kind of harmbench of compliance rates nor any kind of baseline of before and after. Do you think people who uncensor models don't show the benchmark or compliancy scores or any kind of scores on accident? Do you think people who know how to uncensor models don't know or think of to benchmark their model?
People who "uncensor" models but then don't show any kind of scores or tests or empirical numbers know very much what they are doing, and know very much why they aren't doing it. It upets me because making people waste time to download a model in which they have no real expectations of how it should perform is kinda messed up. Some people have bandwidth limits, making them download a model and waste time and energy only for them to find out it isn't very coherent isn't very nice. Low transparency is not cool.
Doing a comparison would be directly punching downwards, so no I'm not going to do that.
I actually wouldn't recommend using this model either. I'm actually considering removing all of my non jang models.