Hello,
I have discovered, that by default the "num_predict" parameter is null, cauinsg Ollama to use its default of 128 tokens. The Ollama API itself allows a value of -1 for unlimited generation, but negative values are not allowed in the AI Studio's MaxTokens parameter. The result is also harder to process, because "done" and "done_reason" values are not universally processed and they are only available in the raw data.
Thanks for your attention.
Regards from Prague!
Jaroslav