Issue Report with GLM-5.3-Flash-UNCENSORED-FP8-UD-Q4_K_XL

[IllIIllIlllIIl]

Hi team,

I ran into a crash while using Koboldcpp and would like to report this issue.

I don't have a technical background, so I'm not sure what went wrong under the hood. However, I managed to capture a screenshot of the console window right before it crashed.

Here are the details of my setup:

• Model format: GGUF
• Koboldcpp version: 1.20
• Operating System: Microsoft Windows 10 Pro, v10.0.19045/ NVIDIA RTX Enterprise 616.92
• CPU: AMD Ryzen Threadripper 7960X 24-Cores
• RAM: 256.0 GB
• GPU: GPU 1: NVIDIA RTX A6000 / GPU 2: NVIDIA RTX A6000 / GPU 3: NVIDIA RTX A6000 / GPU 4: NVIDIA GeForce RTX 4090

Thanks for your help!

[henk717]

This unknown architecture error means the model is not supported by the version of KoboldCpp you are using.
You can use https://koboldai.org/rolling to get earlier access or wait for a version of KoboldCpp that has support.

[henk717]

In addition llamacpp does not support this model, so we can not add support for it at this time rolling won't have it either.
Once ggml-org#27754 is part of llamacpp we will get support in the koboldcpp release after we sync up to upstream. This means its unlikely to be in the next version unless it merges very soon. But i'd assume it will land in the version after that.