Hi everyone,
during my tests of the Frontline Copilot Trigger functions occasionally the trigger failed and returned an error message which mentions that I should ask my Tulip admin for help. When I reduce the size of the input (e.g. images with less file size or less table rows) the trigger works as expected.
Could you please provide limits regarding the input size (e.g. file size or amount of signs) since there are no such limits mentioned in the blog posts? You might also improve the error messages in this case.
Kind Regards,
Oliver.
Hey @OliverGnepper -
Right now, this action has a limit of 2500 tokens, which is about 10,000 text characters on its input. We are seriously considering upgranding this feature to support 30x this limit as we begin to transition this feature out of beta, but we dont yet have a solid timeline on this just yet.
Historically this was driven by the models we are using, but in the last year, these models have significantly improved in their ability to take in a lot of data.
Importantly the operator chat widget is the direction we are moving to suppor the query of signifncant datasets. This widget is trained proactivly, so user questions will be very performant, even with thousands of rows of data.
Pete
Hi @Pete_Hartnett,
thanks for your fast response. In case of the operator chat widget I would appreciate an option to retrain the widget, when the table data changes. Currently, it seems like I have to train it with another table and retrain it again with the original table. Furthermore, I found i quite challenging to retrieve data which is spread across several rows and columns. So I would appreciate it, if you could share some guidelines regarding the data preparation to retrieve information which is distributed over rows and columns consistently.
Kind Regards,
Oliver