Models with Covariates Support
These models can incorporate external variables (covariates) to improve forecasting accuracy.| Model | Description | Use Case | Model ID | Paper |
|---|---|---|---|---|
| Lagllama | Foundation model from Mila Quebec | Best for general forecasting tasks, financial data, and scenarios where external variables significantly impact outcomes. Leverages transformer architecture for complex pattern recognition. | forecast-model-1 | Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting |
Models without Covariates
These models work with univariate time series data and are optimized for specific use cases.| Model | Description | Use Case | Model ID | Paper |
|---|---|---|---|---|
| chronos-bolt | Amazon’s lightweight Chronos model | Best for real-time forecasting and IoT sensor data requiring low latency predictions. Provides fast inference times with minimal computational overhead. | forecast-model-2 | Chronos: Learning the Language of Time Series |
| TiRex | NXAI’s model based on xLSTMs | Best for complex seasonal data and long-term forecasting with intricate cyclical patterns. Designed to capture complex temporal dependencies. | forecast-model-3 | TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning |
| TOTO | Datadog’s observability-optimized model | Best for DevOps monitoring, infrastructure metrics, and application performance data. Tailored for observability-related forecasting needs. | forecast-model-4 | TOTO: Time Series Optimized Transformer for Observability |
More models coming soon!
Model Performance
Each model is optimized for different scenarios:- TabPFN-TS: Best overall performance for complex forecasting tasks
- chronos-bolt: Fastest inference times for real-time applications
- TiRex: Superior performance on seasonal and cyclical data
- TOTO: Optimized for observability and monitoring use cases