Tabby enables versatile forecasting and detection from long time series
Tabby: An Open Pretraining Recipe for Time Series Foundation Models
Machine Learning
Summary
Time series data, like stock prices or sensor readings, often have many points that machines need to understand for tasks like predicting the future or spotting unusual patterns. The authors created Tabby, a large AI model trained on a wide variety of real and synthetic time series data to handle long sequences and multiple tasks. Tabby uses a special Transformer design and learning process to improve performance without changing its core after training. It works well without extra training on new datasets for forecasting, classifying, and detecting anomalies.
What this means in practice
- •For data engineering teams: Provide a pretrained backbone that supports forecasting, classification, and anomaly detection with long time series data across various domains without additional retraining.
- •For financial analysts: Use a prompt-tuned forecast model on long market time series for improved prediction accuracy without retraining the entire model.
Authors
Shifeng Xie, Bahaeddine Abdessalem, Zehao Xiao, Youssef Attia El Hili, Ambroise Odonnat, Zhiwei Dong, Lei Zan, Themis Palpanas, Jianfeng Zhang, Lujia Pan, Keli Zhang, Malik Tiomoko
Abstract
In this report, we release Tabby, a long context probabilistic time series foundation model, together with a complete and open recipe of how it was built. Tabby adopts an encoder-only patch Transformer architecture and concentrates the contributions on the data and the training procedure. The pretraining corpus combines an extended real-world collection, GIFT-Eval-Pretrain+ and BLAST, with synthetic data from KernelSynth and CauKerV2, an online generator that composes temporal dynamics through randomly sampled structural causal models. Training couples a progressive convergence schedule, which yields reusable intermediate checkpoints, with a deep quantile supervision objective for intermediate layers. The resulting 145M parameter backbone supports contexts of up to 8,192 observations and serves forecasting, classification, and anomaly detection, while a prompt-tuning module further improves in-distribution forecasting performance with the pretrained weights frozen. Tabby achieves competitive zero-shot forecasting performance on GIFT-Eval and the out-of-distribution TIME benchmark, while the same pretrained backbone also supports classification on the UCR Archive and zero-shot anomaly detection on TSB-AD-U. We release training pipeline and model as open source at huawei-noah/trustworthyAI.