Take the following as an example. We train a model with different configurations for 1000 steps. Then, the Runs Table only shows the final values of the logged metrics. In some cases, we might want to compare different runs within 500 steps. Can we add support for this functionality?
Hi @lanlin thank you for your question! This is indeed not feasible today, and I can make a feature request if you’d like. You could get the intermediate values for all Runs using
Weave in your Workspace instead. An alternative option would be to download the metrics you’re interested at and log a new column using the API for the specific step, then you could display this in the Runs Table section. Would any of these work for you? Let me know if you would like some example of these workarounds.
Hi @thanos-wandb thanks for your reply. For now, I use the API to upload a new column to the Table. But it would be better to allow some operations directly in the web UI.
Hi @lanlin just an update on this, I have created a feature request for the engineering team as this would be indeed a very useful addition. Thanks for this suggestion, we will reach out to you here if there are any updates on this.
In the meantime, another alternative would be to query these results from a Weave panel that you can add in your project’s Workspace and provide an expression such as:
runs.history.concat.filter((row) => row["epochs"]==500)
I hope this helps, please let me know if you have any further questions.