Datadog Releases Open-Weights Toto 2.0 Time Series Foundation Models on Hugging Face

Datadog has unveiled the Toto 2.0 weights in Apache 2.0 on Hugging Face, introducing a family of open-weights time series foundation models (TSFMs) ranging from 4 million to 2.5 billion parameters. Each model size outperforms the previous one using a single hyperparameter configuration, marking a significant advancement in the field. Toto 2.0 has achieved top scores across major benchmarks, including BOOM, GIFT-Eval, and TIME. This development suggests that scaling laws, which have previously proven effective for language and vision models, are now applicable to time series models, allowing for confident scaling of data and compute resources. The open-source weights can be accessed at Hugging Face.