Issue with Phi Data Framework: Token Exhaustion and Agent Interruption

API
1

Hi,

Issue: I am using the Phi Data framework to build agents and have encountered an issue.

When I call my agent, it invokes a certain number of tools and receives a response from an API. However, the response is too large, and after a certain point, I receive an error message: “Token Exhaust 429”, “Force Reason = RECITATION”, or “Force Reason = Stop.” As a result, my agents stop working midway.

Can anyone explain why this is happening? It seems like the issue is related to output tokens being exhausted, but when I checked, the total output tokens were around 6,000, whereas the maximum output token limit set by me is 10,000. My temperature is set to 0.3, and I am using the Gemini 2.0 Flash model.

Any insights would be greatly appreciated!
Thanks

2

“recitation” is an output termination error type when the AI seems to be reproducing direct from its pretraining data or output is detected by a copyright system.

You’ll need to complain about Google AI model performance issues in their own support platform:

1 Like