finetuning univariate forecast worse zero-shot, why?

[m6129]

Dear developer. I would like to draw your attention to the fact that at the moment, finetuning on datasets with the purpose of univariate forecast, is worse in terms of forecast accuracy than zero-shot forecasting