Hi there, I'm currently facing a performance issue during my model inference. When measuring the requests, I observe that after approximately 5 minutes, the first request slows down. Here are the elapsed times I recorded for two rounds of 20 requests each:

** 1 round, req. 1 ** Elapsed time: *0.7954* seconds
...
** 1 round, req. 20 ** Elapsed time: 0.1974 seconds
Sleeping 360 sec to simulate cooling down and warming up ...
** 2 round, req. 1 ** Elapsed time: *0.6992* seconds
...
** 2 round, req. 20 ** Elapsed time: 0.1885 seconds

In my tests, I hit the API with the same data for 20 times in succession and measure the time taken for each request. Some additional details: - Model Details: I'm using a basic sklearn model with stacked classifiers. - Environment: local machine - Payload Size: small text up to 40 characters I'm wondering if anyone has encountered a similar issue or has any suggestions on optimizing this. Specifically, any tuning or configuration tips for improving the initial request time after a break? Thanks in advance for any insights or guidance!