Search code examples
google-cloud-platformgoogle-cloud-vertex-aigoogle-cloud-automlvertex-ai-pipeline

Vertex AI - Training Performance Graph


I'm looking for a graph on Vertex AI that can show me around which time the performance converged so I can present it & set the future trainings' budget (I would like to stop before the automatic early-stop).

On 2 different articles on Vertex AI, I saw a "Training performance" graph (as below). In this medium article and an another in-company article written by an employee who left the company since. I see no directions to access this graph and no mention of it anywhere else except an answer to this question which suggests it doesn't exist 🤔

There is also this question which is about monitoring from logs, but I'm convinced that a graph exists somewhere.

How do I see this Training performance graph, under which menu is it?

enter image description here

enter image description here


Solution

  • Thanks to the help from the writer of the Medium article, we were able to discover:

    This graph is visible in the training details page, which is easily accessible by clicking on the training while under "Trainings" menu.

    However, after a training is completed, clicking on it keeps redirecting to "Model Registry" menu, which doesn't show the graph.

    I had to force my completed training's ID into training details link and see the Training page along with the graph.

    https://console.cloud.google.com/vertex-ai/locations/{LOC}/training/{TRAINING_ID}

    enter image description here