Is anyone able to run gemma-2-9b-it-WPO-HB-GGUF (Q8_0)?

[Hardeh]

Link to model: https://huggingface.co/mradermacher/gemma-2-9b-it-WPO-HB-GGUF
I'm using SillyTavern, koboldcpp_rocm and 7900XTX GPU with 24GB VRAM. I can run many models, for example i can run q8_0 quants of 12b models based on mistral nemo, fully on GPU, with 32k context. Or Llama 3 8B models as well. But this model takes up all GPU memory with 32k context, and with 16k context it takes about 18 GB of VRAM, and runs so extremely slowly that it's unusable. (0.18T/S).
With 8192 context, model takes up about 15 GB of VRAM, and still runs as slow.
What can be reason of this? I see something like this happening for the first time, every other model i tried seems to run just fine.