VRAM overfill between files when batch processing

[Jax2025]

When I process 1 video at a time, I have no issue with delays or slowdowns from going over the dedicated GPU memory into shared GPU memory - however, when I process an entire folder, the program will begin processing the next video before the VRAM usage has time to be lowered. This leads to situations where the VRAM usage stacks up the two, and the next video goes from rendering at 60fps to 1-2fps. At no point after that does the VRAM usage lower, it will stay like this until the entire batch is done. Is there currently a solution to this? Love the software, this is the only issue I've had with it. For reference I am using Any_V2_S, Full SBS, 4 depth batch size and 4 worker threads, fp16 and stream on an Nvidia 4070.

(The only temporary solution I've found is to just cancel the processing and re-start it by hand, but this defeats the purpose of being able to batch process it for me)

[nagadomi]

Not sure why python's auto garbage collection is not working,
but I updated the code to explicitly release VRAM for each video processing.
Try again after running update.bat.

On my Linux PC this problem does not happen, so I don't know if the problem has been fixed.

[Jax2025]

That does appear to have fixed the issue! Thank you so much, you do amazing work and the fact that you were able to get it fixed so quickly is outstanding [:

[nagadomi]

Thank you for the confirmation.
I did a little digging and it seems that python's garbage collection may not work well because it doesn't monitor VRAM usage.

[diverswan9]

This is so cool! This is exactly the same problem I've had on my 4090 but not on my 3060. I will try this out. Cheers!