I recently saw a new feature in AIP logic under input call Add interface. When you click on it it shows up the below screen
I wasnt able to create variables as per the error message. There isnt any documentation on how to use it.
I think this is supposed to provide ways to use Vision based features of the LLM, but not sure how to use them.
Function interfaces define constraints on a function’s inputs and/or outputs. This enables other applications to automatically pull in and use compatible functions once published (think function pickers in other apps that need specific functions).
I wasn’t able to create variables as per the error message. There isnt any documentation on how to use it.
Which error message are you getting?
I think this is supposed to provide ways to use Vision based features of the LLM, but not sure how to use them.
The particular contract you have highlighted allows a function to be detected as a vision model.
You don’t need to add an interface to use vision though, you just need:
Then you can reference the media property in the prompt and it will pass the image to the LLM.
may be im not asking the right question. Can you provide some examples on how to use function Interfaces in AIP Logic ?
If you are trying to pass media to a model that supports vision you do not need to use interfaces, you should remove the interface and pass the media reference into the “Use LLM” block’s prompt.
The particular interface you have selected doesn’t look like it should be causing validation errors though so we need to investigate and fix.
Update:
Yes, those two parameters required by the interface are actually structs:
What would be the functionality of these interfaces ? I dont see explanation of how to use it in AIP logic documentation. Do I need to use these when using “Bring your own Model” ?
Here are the public docs for Function interfaces:
https://www.palantir.com/docs/foundry/functions/function-interfaces/
There isn’t any Logic specific documentation at the moment
Hope you can soon add the other two intefraces
chatcompletionwithVision
embedding
