[GENERAL SUPPORT]: performance impact of `max_trials` in `get_next_trials`

[zhqrbitee]

Question

Hi, I'm looking at the max_trials parameter on Client.get_next_trials and wondering about its performance impact. If I set max_trials=4 instead of 1, should I expect throughput to improve — say, by amortizing the per-call overhead down to roughly 1/4?
In the old AxClient.get_next_trials, I believe the answer is no, since it is just a simple loop that calls AxClient.get_next_trial. But I'm not sure whether the new Client class introduces any batching or parallelism that would actually improve throughput here.

Please provide any relevant code snippet if applicable.

Code of Conduct

• I agree to follow this Ax's Code of Conduct

[saitcakmak]

Hi @zhqrbitee. In terms of pure candidate generation, it is also using a loop under the hood. So, if you're just generating multiple candidates without attaching any new data in between, the two should be practically equivalent.