Query are slooooooooooooooooooooooow

[paciox]

WrenAI latest. local docker setup with postgresql where I pull data from
Linked it to a Openrouter model.
It's SLOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOWWWWWWWWWWWWWWW

it stays there for 30 minutes an
I managed to run it only once and never managed to do charts:

how i can diagnose?

[cyyeh]

@paciox are you running on gpu?

[paciox]

@paciox are you running on gpu?

Yes.
Turned out Openrouter is not the best choiche.
Wren fires multiple concurrent queries and hogs Openrouter that simply doesn't want to hold that much traffic all at once.

Ran some local LLM on LMStudio and managed to sort out both queries and charts.

That said, compiling the configuration file has been very painful and complicated and still, it's quite slow as a tool. It takes a good 2-3 minutes to get all the data out.

Will have to test it out with higher token per minute models or different techniques.

[JackieLieu]

don't use thinking model