[Question]

[northvip]

In co-storm, expert responses usually require significant delays. Apart from large model requests, which stage takes the most time? I want to optimize the delay in this part.

[shaoyijia]

Please check out the entry file of STORM, it decouples the full pipeline into multiple stage in run(). You can add timer to profile each step there.