Skip to content

Conversation

AyushExel
Copy link
Contributor

@AyushExel AyushExel commented Oct 13, 2021

Addresses #5160

πŸ› οΈ PR Summary

Made with ❀️ by Ultralytics Actions

🌟 Summary

Improved Weights & Biases (wandb) logging integration in YOLOv5.

πŸ“Š Key Changes

  • πŸ“¦ Added pkg_resources (as pkg) for version checking.
  • πŸš€ Modified wandb login process with version check and a login attempt within a conditional block.
  • πŸ’‘ Moved wandb version check and login from wandb_utils.py to __init__.py within loggers.
  • 🌐 Ensured wandb login only occurs on the main process (when RANK is 0 or -1).

🎯 Purpose & Impact

  • πŸ“ˆ Ensures better package version management for wandb, avoiding potential incompatibilities.
  • πŸ”’ Improves overall stability with a more controlled wandb login flow.
  • πŸ›  Users will experience more reliable wandb integrations, especially in distributed training scenarios.
  • ⏱ Adds a timeout to wandb login for better failure handling.

@glenn-jocher
Copy link
Member

@AyushExel I think this PR might have introduced a new error. If I pull the latest image including this PR, run wandb login API_KEY, and then train normally, I get the following error:

Screen Shot 2021-10-13 at 6 37 59 PM

@Chigedaxigua
Copy link

@AyushExel I think this PR might have introduced a new error. If I pull the latest image including this PR, run wandb login API_KEY, and then train normally, I get the following error:

Screen Shot 2021-10-13 at 6 37 59 PM

I got the same issue today, did you slove this NameError yet?

@Chigedaxigua
Copy link

thank you!

@AyushExel AyushExel mentioned this pull request Oct 14, 2021
@AyushExel
Copy link
Contributor Author

@glenn-jocher ahh yes. wandb needs to be explicitly imported. Fix #5176

BjarneKuehl pushed a commit to fhkiel-mlaip/yolov5 that referenced this pull request Aug 26, 2022
* fix dpp with wandb disabled

* PyCharm reformat

Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Docker Multi-GPU DDP training hang on destroy_process_group() with wandb option 3
3 participants