MOMENT
Encoder-only family of open time-series foundation models, pretrained on the Time-Series Pile. Building blocks for forecasting, classification, anomaly detection, and imputation; effective zero-shot and tunable with light task-specific data.
MOMENT (Auton Lab, CMU, ICML 2024) is one of the first openly released general-purpose time-series foundation models. The architecture is an encoder-only transformer trained with a masked-reconstruction objective on the “Time-Series Pile” — a large, diverse, public time-series corpus assembled by the authors specifically to enable large-scale multi-dataset pretraining.
The released checkpoints are designed as building blocks for multiple downstream tasks: forecasting, classification, anomaly detection, and imputation. They work zero-shot, with few-shot adaptation, or with full fine-tuning. TS-Arena runs the Small / Base / Large MOMENT-1 checkpoints.
Versions on TS-Arena
Each version below corresponds to one registered model id in the leaderboard. Click through to its detail page for per-model rankings, forecasts, and history.
- MOMENT-1 Smallmoment-small40M params…
Smallest MOMENT-1 checkpoint.
- MOMENT-1 Basemoment-base-model125M params…
- MOMENT-1 Largemoment-large385M params…
Largest MOMENT-1 checkpoint.