Toto 2.0: Time Series Forecasting Enters the Scaling Era
TL;DR AI
2 min readKey summary
Researchers introduced Toto 2.0, a five-model time series forecasting family built with one unified training recipe.
The models scale from 4M to 2.5B parameters and improve consistently as size grows, showing strong scaling behavior.
Toto 2.0 set state-of-the-art results on BOOM, GIFT-Eval, and TIME.
Datadog released the base checkpoints and training approach under Apache 2.0, making the models openly usable.
