Quantcast
Channel: Configuration - Home Assistant Community
Viewing all articles
Browse latest Browse all 105521

Anyone else have issues with their LLM just....not calling an intent?

$
0
0

Having a weird issue whereby Gemini Flash 2.0 appears to often just…not call an intent.

I suspect there’s actually more to it than that and it IS calling it, albeit perhaps wrongly, and there’s just no log of it - or it’s calling it correctly and it’s being dropped.

With debug logs for google cloud, components.conversation and components.google_generative_ai_conversation I get the following (there’s nothing after this):

Summary (click for more details)

Notably weird is the lack of any evidence of an intent call. This seems to happen fairly randomly.

edit: perhaps it’s when there’s already a previous, historical event call in the history?

I’ve tried altering the initial prompt to be really clear about always calling the intents accurately, and I’ve tried talking it through using them “Use HassTurnOff to turn off the study light…” and it still seems to randomly not work. The first time always works, though.

I also find it weird that the LLM always sends back some text along with the intent request that seems to be discarded, and then sends back some more text after it gets the confirmation - surely this causes lag as there’s two round trips and two lots of processing?

1 post - 1 participant

Read full topic


Viewing all articles
Browse latest Browse all 105521

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>