Open WebUI

Objective

In order to facilitate AI use and allow users to experiment with different models and tools, we host an instance of Open WebUI at https://aichat.phys.ethz.ch. You can either use external models or request an API token for our onprem models. In both cases, you'll have to tell Open WebUI about the models you'd like to use.

How to configure our onprem model in Open WebUI

Here we describe how to add our onprem Qwen3.8-27B-FP8:

  • Click on your avatar and open Settings
  • Click on Connections
  • Click on the plus sign near Manage OpenAI API Connections
  • In URL, add https://aimodels.phys.ethz.ch/v1 and in Bearer add the sk-xyz token you received from us
  • Save
  • Your browser will then ask you if you'd like to allow local network connections. You'll have to agree to that.
  • There's one more setting you'll have to change in order for auto compaction to work correctly: Settings -> Models -> Qwen3.8-27B-FP8 Private -> ✏️ -> Advanced Params -> Context Compaction Threshold -> enter 150000 -> Save & Update

At the very bottom of your chat window you can now choose Qwen3.8-27B-FP8 as your preferred model. Don't use Qwen3-Embedding-4B as this is for dedicated embeddings.

Notes

As the LLM requests are coming directly from your browser and out GPU hardware is in an internal network, aichat.phys will only work within the ETH network or via VPN.