Maximum Output Tokens from Tuned Models

Google AI Studio
,
1

Hi All,

Currently I am using Gemini 1.5 Flash tuned model for my use case and I noticed that it does not return more than 1000 tokens (4000 Characters). I read multiple posts on this forum where people mentioned that maximum output tokens are around 8000 while some people mentioned receiving around 1800 max tokens in output. Any idea how I can achieve higher maximum output tokens? For my use case, even 1500 output tokens are sufficient.

Thanks,
Kapil

2

Question: when training your tuned model, what was the average size of the “expected answer” column in your training set? The conjecture is, you might have trained it to give brief responses.

4

Thanks for your reply.

The model was trained on around 5000 characters which represents a complete script (Something that has to be delivered to user) that model is supposed to generate. Now when executing the prompt the output characters are around 4000 with incomplete script. It means that model did generate the complete script like it is trained on but it returned less number of characters and truncated the remaining script which is causing my APP to deliver incomplete content to users.