| Property | Value |
|---|---|
| Parameter Count | 311M |
| License | CC-BY-NC-4.0 |
| Paper | arXiv:2402.02592 |
| Author | Salesforce |
| Model Type | Time Series Forecasting Transformer |
What is moirai-1.0-R-large?
Moirai-1.0-R-large is a sophisticated time series forecasting transformer model that represents the largest variant in the Moirai family. It's a masked encoder-based universal time series forecasting transformer pre-trained on the LOTSA dataset, designed to handle complex time series predictions with remarkable accuracy.
Implementation Details
The model utilizes an advanced architecture that processes time series data through patch embeddings, combining sequence and variate IDs within a transformer framework. It supports multiple variate time series and can handle both target variables and dynamic covariates.
- Flexible patch size configuration (8 to 128 or auto)
- Support for multiple prediction lengths
- Customizable context length for historical data
- Batch processing capabilities
Core Capabilities
- Universal time series forecasting across various domains
- Handles multi-variate time series data
- Supports dynamic covariates in forecast horizon
- Generates probabilistic forecasts with mixture distributions
- Efficient processing through patch-based tokenization
Frequently Asked Questions
Q: What makes this model unique?
Moirai-1.0-R-large stands out for its universal approach to time series forecasting, leveraging a large-scale architecture with 311M parameters and a novel patch-based embedding system that can handle various types of time series data efficiently.
Q: What are the recommended use cases?
The model is ideal for complex time series forecasting tasks requiring high accuracy, particularly in scenarios with multiple variables and known dynamic covariates. It's suitable for business forecasting, demand prediction, and other time-dependent predictions requiring robust probabilistic outputs.