Hi Team,
I’ve encountered an issue with Gemini 2.0 Flash Thinking Experimental 01-21 and Gemini Experimental 1206 when generating responses in Chinese. The models often stop outputting mid-sentence and, in some cases, return an An internal error has occurred message instead of completing the response.
This issue happens every time when generating text in Chinese. It doesn’t seem to occur when generating text in English.
Could you please investigate this bug? Let me know if any additional details or logs are needed.
Thanks!
13 Likes
Good morning I would like to suggest to keep in mind that there are two different dialects of Mandarin and when you are using sentences in English or French or whatever language you choose when the model has to convert that into Mandarin there is this possibility that the dialects could play a role in the stopping of the response mid sentence have you perhaps tried specifying which dialect, what precisely is the model conversing about there are some instances where the explanatory constructs may have conflicting conversive and expressive responses