UVR is not freeing VRAM after inference

[Zcooger]

As seen by some users.
I noticed it personally after using V4 models with stacking and not.

[DilanBoskan]

I will look into it.
I guess a temporary work-around would be to restart the application after an inference.

[Zcooger]

Good, once application is restarted keeping last selected models will be useful too.

[DilanBoskan]

Can you give me more information? How much VRam is being used up, is it only on v4 or also on v2? With which options is the VRAM usage the highest?

As seen by some users.

By who and by how many?

Every detail helps, thank you.

[DilanBoskan]

Please try out the newest commit in the dev branch.

[Zcooger]

Tested and working properly, VRAM is now cleaned after inference.
From total 1,98GB VRAM during inference regardless of settings 0,23GB is still free. 1,4GB is free in idle.
3 or 4 users of Audio Separation Discord that use UVR GUI noticed that.
Behavior of that was really random.

[DilanBoskan]

Thank you for the info
Marking closed.