NX-AI/tirex

What it solves

TiRex is designed to provide high-accuracy time series forecasting without requiring the user to train the model on their own specific dataset. It addresses the challenge of creating reliable predictions across both short-term and long-term horizons using a pre-trained approach.

How it works

Based on the xLSTM architecture, TiRex is a 35M parameter model that utilizes enhanced in-context learning to perform zero-shot forecasting. It can generate both point estimates and quantile estimates for future values. Beyond forecasting, the model can also be used for time series classification and regression tasks.

Who it’s for

Data scientists and researchers working with time series data who need immediate, high-performance forecasting, classification, or regression without the overhead of training a custom model from scratch.

Highlights

• Zero-Shot Capabilities: Predicts future values without needing training on user data.
• Versatile Outputs: Provides both point and quantile estimates.
• Broad Application: Supports forecasting, classification, and regression.
• Optimization: Available as an ONNX model for optimized inference and supports custom CUDA kernels for improved performance on compatible GPUs.