How do I tag/label multiple runs?


I’m struggling to achieve the following:
In TensorBoard I can only differenciate multiple runs by ID, as the comment is truncated, so the hyperparameters are mostly not shown.
For example:

42afef82 my_model:train 2022-08-23 02:02:04 architecture=DenseNet-121 batch_size=24 classes=14 comment='DenseNet/experiments/lightning_logs/version_0

In this example, the hyper-parameter dropout is not visible.
Now, what I would love to do is to compare specific runs, e.g. all runs with dropout=0.1, in TensorBoard.

Based on the docs AFAIK the only possibility is to filter the runs by tag and then run TensorBoard specifically for these runs, instead of using guild view → click on TensorBoard → filter inside TensorBoard.

Therefore, my question is: how do I tag/label multiple runs?

I discovered, that I can display all dropout=0.1 runs using:

guild runs list --list-all -Fl=dropout=0.1

However, running guild runs tag -Fl=dropout=0.1 -a dropout=0.1 tags only one run.

Furthermore, I also want to tag runs where dropout wasn’t a hyper-parameter but hardcoded. How do I select those? Is there something like a negative lookaround?



Hi Alessandro,

I can’t try your example because I don’t have your experiments.
However, I think you can add ‘:’ at the end of your command as in:
guild runs tag -Fl=dropout=0.1 -a dropout=0.1 : to tag multiple runs.



Awesome, that works. Thanks Josue!