Rakuten W&B Onboarding Material



How to start W&B?

Please go through this page!
Please set up your environment variable WANDB_BASE_URL as https://rakuten.wandb.io/ 
W&B Models demo & hands-on
Script: https://colab.research.google.com/drive/1KyiRmcFAu-dk0pA2PYo8oHlEKldvzqqj
Material: https://docs.google.com/presentation/d/1oJi-SKE35CXT9JieEGev6e9wdWawHtzFzqnoe3GupqQ/edit?slide=id.g2bd3e21e6d8_0_0#slide=id.g2bd3e21e6d8_0_0
W&B Weave demo & handso-on
Script: https://colab.research.google.com/drive/1yEGr67XE2lon0wm_8evPnEVBFj2HhHXS

What we will achieve in PoV

Training generative models (image and video generation)
Training discriminative models (object detection/image classification)
Validate the generated output scores



Reference

List of differentiators against mlflow https://docs.google.com/presentation/d/1g7wsCXUCYOFIj3XEvfjInmSS7CDWDQ6fw-wZbNT6xZc/edit?usp=sharing
How to show the diff of texts in W&B Table: https://www.loom.com/share/5f8a70bee88642c5a1d82ab9b5700cd8?sid=832f878c-39fd-4f4a-b123-c9d6c2f2460f

Support for Training generative models (image and video generation)

✅ Check videos over time

✅ Compare validation results of a specific image over time

✅ Compare the same validation sample across different runs side-by-side

💪Customize time step axis in Tables (e.g. sync with `trainer.global_step`) → Product Development

✅ Best practices for multi-GPU or multi-node setups with W&B**

✅ Check tables in the middle of evaluation step