Abrupt Ending Token

In a lot of google gemini models, especially when coding, the model seems to predict the EOF token randomly, like at float endings like f0.6 as if that was the end.

Hello @siriustk ,

Can you share the following…?

  • What platform you are using, Gemini API or AI Studio?
  • Which models have you tried this with?
  • Could you give an example of a code snippet where this occurred and the prompt you used to generate it?