By default max output tokens are 16. It must not
In chat mode it is not counted when this option is hidden. Here you can see translation to Russian which usually takes 1 token per character. Obviously it’s more than 16
If I show tokens option but not change it, their number is counted:
For Text completion, there’s no bugs. Even when Max tokens option hidden, it’s still in power
The API default for OpenAI is 16 so if it isn’t set it will use what they define, we have an internal feature request to change the default for this at some point
@Jon Well, then it’s not clear what in n8n means “hidden” properties. Does it mean they are not sent? Or n8n sends the value which you can see by showing the option?
Need to understand it for my course. I thought the latter.
Nevertheless, in OpenAI code different resources take this differently.
Additional options are either set with a default we send or for some nodes like OpenAI they are not sent at all and the api controls the default. If you add the additional option it will be sent with the value you input.
If you check the API docs for OpenAI you can see that Chat Completion no longer has a default max limit and text completion is still 16.
Let me know if I have missed anything else on this one.
@Jon Thank you. Will consider hidden options as “it depends on API”
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.