Olmo was pretrained on monthly aggregates, but this looses information at finer time-scales. However since olmo was trained on sequences of length 12 (the monthly aggregates) I wonder is there are any tradeoffs moving to a more dense time series, e.g. length 52 for an imager per week. It would be useful to hear (1) if this has been investigated and (2) is there is any intuition about the impact of a dense time series.
Thanks
Olmo was pretrained on monthly aggregates, but this looses information at finer time-scales. However since olmo was trained on sequences of length 12 (the monthly aggregates) I wonder is there are any tradeoffs moving to a more dense time series, e.g. length 52 for an imager per week. It would be useful to hear (1) if this has been investigated and (2) is there is any intuition about the impact of a dense time series.
Thanks