About TimeGPT
Get started with our QuickStart guide, walk through tutorials on the different capabilities, and learn from real-world use cases in our documentation.
Architecture
Self-attention, the revolutionary concept introduced by the paper Attention is all you need, is the basis of this foundation model. TimeGPT model is not based on any existing large language model(LLM). Instead, it is independently trained on a vast amount of time series data, and the large transformer model is designed to minimize the forecasting error.
The architecture consists of an encoder-decoder structure with multiple layers, each with residual connections and layer normalization. Finally, a linear layer maps the decoder’s output to the forecasting window dimension. The general intuition is that attention-based mechanisms are able to capture the diversity of past events and correctly extrapolate potential future distributions.
To make prediction, TimeGPT “reads” the input series much like the way humans read a sentence – from left to right. It looks at windows of past data, which we can think of as “tokens”, and predicts what comes next. This prediction is based on patterns the model identifies in past data and extrapolates into the future.
Explore examples and use cases
Visit our comprehensive documentation to explore a wide range of examples and practical use cases for TimeGPT. Whether you’re getting started with our Quickstart Guide, setting up your API key, or looking for advanced forecasting techniques, our resources are designed to guide you through every step of the process.
Learn how to handle anomaly detection, fine-tune models with specific loss functions, and scale your computing using frameworks like Spark, Dask, and Ray.
Additionally, our documentation covers specialized topics such as handling exogenous variables, validating models through cross-validation, and forecasting under uncertainty with quantile forecasts and prediction intervals.
For those interested in real-world applications, discover how TimeGPT can be used for forecasting web traffic or predicting Bitcoin prices.