Got this feedback from a user:
“There’s been probably 15 times where it doesn’t fully load its response. It will do half or less and then just stop, but it counts as a credit.”
Any tips here?
Hi @femalenatured,
Could you provide some examples of the incomplete responses occurring in your workspace? Any sessions ids, users, or time frames where this was happening would be especially helpful! A quick suggestion that I can give you is increasing the output token slider for your agent. Send me the info so we can get your issue resolved soon!
thank you for that feedback right now my output token length right now is 5000 if that means anything.attached is a screenshot from the user who was having trouble’s message log that I can see not sure if this has anything to do with it but seems like outputs taht weren’t generating? thank you
Hi @femalenatured,
We are looking into your issue right now. I’ve created a ticket for you under PRD-1239 and I’ll give you an update when I get word! It seems like this issue is more about the agent not responding at all rather than loading half of a response. If you have any session ids where the agents is only loading half the response, please send them over!
Hi @femalenatured,
We have finished evaluating the issue described in PRD-1239. It looks like the error that you showed here is a streaming error on Gemini’s end. The fix we made was to ensure that a fallback model would be used if enabled. If fallback models are toggled off, the response would end as an error. Let us know if your users continue to experience this issue!