I've been using Prodigy for annotating NLP models without any issues the past 6 months. Only a few days ago I started having trouble with the UI getting stuck in Chrome. I'm trying to label a spancat model with 4 labels and only 1800 samples. I can label 60 or samples then when I press the green OK button the page border is highlighted in green and the UI doesn't do anything.
I've tried on another computer and I've tried with a new clean Sqlite database but the same thing happens every time. Even running with verbose logging, nothing shows up in the lo and no errors appear in the Chrome developer console. The UI is simply not working. I'm using version 1.11.14 with Python 3.9.15
Any ideas what might cause the green border and lack of response from the UI?
Is it possible that this is related to a Chrome update?
Does the issue persist in other browsers as well? Firefox?
Could you share a GIF of that you're experiencing? That might help me debug.
I can label 60 or samples then when I press the green OK button the page border is highlighted in green and the UI doesn't do anything.
Just to check, hitting "accept" does not send any data back to the server yet. Data is only sent back once you have a full batch. This is to ensure that you'll able to correct yourself by going back a few examples. In order to force a save, you can either use the keyboard shortcut (CTRL + S/CMD + S) or use this button:
That is most strange. I'd love to try and reproduce this locally. Could you share an excerpt of the dataset? It can also be a dataset that differs from what you're using, as long as it causes the same bug to appear.
Could you also share the logs that show up in the terminal?
One thing, I couldn't help but notice that you are sending data to a non-local IP address. If there are firewall rules on the server that is hosting this then that may cause the issues too. One thing that you might want to double-check is the interactive docs feature. Prodigy uses FastAPI under the hood and you should be able to interact with the backend via the docs. In particular, it might be good to confirm that you're able to send POST requests, just to make sure that these aren't prevented somewhere.
