I have some models on GCP AI platform deployed via custom prediction routine (legacy mls1 type machines), however the performance and resource usage charts are always not available with the following massage "No data is available for the selected time frame".
This is not the case with my other models deployed via custom containers (N1 machines) where i can immediately see the metrics in the charts. I could not find any doc from google about this different behaviours between the two deployment method (or machine type), as it seems to suggests that (according to this google doc) all models deployed should have these enabled.
No data is available for the selected time frame
Does anyone know why there's this difference and whether and how I can enable those metrics for custom prediction routine endpoints?