#Hi !
1 messages ยท Page 1 of 1 (latest)
Apologies for the troubles. This language check was added for the upcoming Origin Trial since we only have tested English for safety.
Based on the community's feedback, we are making a few changes. In particular, we'll let the output through if the detected language is "unknown". This should help scenarios like yours.
For prototyping and the built-in AI challenge, we are also looking into adding a way to bypass the language check.
๐ If possible, please send examples via DM so I can check with the team that the new heuristics will work better.
Hi, thanks for the quick answer.
I'll send you some examples in which I found said issue and the prompt(s) necessary to replicate it by the end of tomorrow (GMT+1).
Good day,
Elias.
Any update on when the detected language thing will be loosened?
Would also be interested, totally unable to build ideas we have simply because for some reason it's seeing the page english as not english enough haha
Would love to see this addressed as well
Hi,
In the end I won't be able to send more examples.
To be honest, I just dropped the project because I saw that the browser-local model is too limited for the use-cases of the extension I'm developing.
A suggestion would be to add an option that would allow devs to load fine-tuned versions of Gemini Nano (from some kind of Google-certified repository, for example), which would put the model on par with 7b or 12b models on application-specific usecases, and therefore much more useful.
Good luck to you guys who participate in the contest !
Hey @silent meadow , I am facing the same issue when trying to get the structured outputs. Example of this can be:
Prompt: Today date is 14 Nov 2024. Derive the date from the Expression "2 days before today" in DD-MM-YYYY format without any explanation or other text.
Error: The model attempted to output text in an untested language, and was prevented from doing so.
Can you please help here! Cheers
I am also facing this issue, did you find any solution?