Open
Description
Hey,
i am running into frequent problems with Whisper WebUI because of the way it handles the VRAM. It seems to reserve most of the VRAM of our RTX3090 and i dont understand if / when it frees the VRAM.
- We are mainly using the large-2 model, which in my udnerstanding does not need ~24GB VRAM, but it seems as if the software tries to grab all the ram it can get.
- When i am running other tools that are using VRAM, Whisper-WebUI seems to crash, instead of reporting an error or trying to continue when RAM is available.
Is there a way i can control how much / when the VRAM is requested?