Low disk space causes memory leak

[oliver408i]

Search before asking

• I have searched the YOLOv5 issues and found no similar bug report.

YOLOv5 Component

Training

Bug

Having only a few gbs of disk space left when running yolov5's train.py causes a memory leak that will cause the computer to lag and eventually kill the yolov5 process. For example, I had ~2 gb of disk space left, and the after a few epochs of training, the "out of application memory" prompt starts popping up, and after a few more epochs, the process is killed by the system for using too much ram.

Environment

YOLOv5 🚀 v7.0-358-gc07b9a8b Python-3.12.5 torch-2.4.0 MPS
MacOS 15. Macbook pro m2 (16 GB ram)

Minimal Reproducible Example

Simply run a decently large dataset (in my case, 300 images) on at least 10 epochs. Using CPU or mps doesn't matter in this case.

Additional

I'm not sure if this is a global problem or just with Macos specifically, as I haven't been able to test it out. I suggest the program to warn the user about low disk space, because the system ram warning made it seem like it was my ram at fault instead of my disk space. Everything works as normal after I cleared up the disk space, using the exact same setup as before.

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[glenn-jocher]

@oliver408i thank you for reporting this issue. Please verify if the problem persists with the latest version of YOLOv5. If it does, we recommend ensuring sufficient disk space before training, as low disk space can impact performance. Your suggestion for a disk space warning is noted and appreciated.

[oliver408i]

Yea I was. I cleared my disk and it works fine now. I just wanted to suggest that a warning be added to avoid other people from spending hours trying to fix this issue only to realize that it was a simple issue. Thanks for responding

[glenn-jocher]

@oliver408i thank you for the feedback. We'll consider adding a disk space warning to help prevent similar issues in the future. If you encounter any other problems, please let us know.