New request
#2083: GPU stays always active after a prompt in the webui
We'll provision a sandbox, run an agent against the issue, and open a draft PR. You can pull the branch and iterate from there.
Running Gwen3.6-35b-a3b on LM Studio on Windows with full context on a RTX 4070 Ti Super. Hermes and the webui running in a LXC container on an ubuntu 26.04 machine in local network.
Using the original hermes chat through the terminal, making an "hello" prompt, the GPU does the job and then stops. However in the webui, same "hello" prompt, the GPU never stops. I have to kill the process.
What can we do?
Terminal chat: <img width="285" height="170" alt="Image" src="https://github.com/user-attachments/assets/e1a04683-27bf-496e-84df-f45c8fad4c83" />
WebUI chat: <img width="287" height="170" alt="Image" src="https://github.com/user-attachments/assets/268d85e7-507a-4cef-b056-d3407298eb91" />
Edit: with a smaller model google/gemma-4-e4b it works better:
Terminal and webui "hello" prompt: <img width="288" height="168" alt="Image" src="https://github.com/user-attachments/assets/6ee94b75-b102-4624-b02c-47d500e60926" />
GPU remaining active after stream completion indicates a missing cleanup or close event in the chat streaming lifecycle.
- api/streaming.py
- api/sessions.py
- api/routes.py
- static/js/chat.js
- static/js/composer.js