WhisperX | Memory leak over time.

[OPPEYRADY]

WSL anaconda Ubuntu | Python 3.10 (Also did the same thing on windows as well it seems)

It seems either when whisperx is ran for extended periods each inference adds a VERY slight increase to current vram, it however seems to keep building and building till eventually it reaches up to 20+ gigs and inference comes to a crawl on my 4090, I only noticed this because of how extensive my program runs in the background.

So far I've been bandaiding around it by restarting the python server over X times or over time to release the memory as other memory releasing commands don't seem to work, but it would be nice if this can be looked into on the development branch.

After around 1000ish or so generations with V3 I started to notice it.