Hi,
I’m currently performing a hyperparameter search (I’m not using wandb’s sweeep feature for that) on a GPU cluster. I group the runs into different categories, let’s say that’s simply “Group A”, “Group B”, etc.
Now, since jobs can crash for various reasons, I would love to have a panel that reports the number of successful runs for each group, so I know how each group is doing (“Group A has 200 successful runs, while Group B only has 50, so I need to start some more jobs for Group B”). I know I can get there by using the general filter and group feature in my project’s run-table, but this is rather tedious and it would be more convenient for me to have it as a panel for quick access (e.g., inside a report).
I’ve fiddled around with the “scalar chart” and the “weave” panel, but without success – any ideas?


