Dataset · thuml
UTSD large-scale time series corpus
thuml/UTSD
A unified time series dataset from 7 domains with up to 1 billion points for foundation models.
What's inside
The Unified Time Series Dataset (UTSD) combines time series from 7 distinct domains containing up to 1 billion time points organized into four hierarchical volume levels. It features time series and text modalities, with a numpy format version available for fast loading. Total record count is indicated between 100k and 1M.
What you can do with it
- Pre-training large time-series foundation models
- Training and testing time-series-forecasting algorithms across domains
- Studying neural network scalability on sequential numerical data
- Tasks
- Time series forecasting
- Size
- 100K–1M rows
- Downloads this month
- 3.8K
- Likes
- 40
Models trained on it
- thuml/sundial-base-128mTime series forecasting 52.9K
- thuml/timer-base-84mTime series forecasting 2.4K
- bytedance-research/Timer-S1Time series forecasting 220
- thuml/Timer-S1Time series forecasting 197
- sktime/Timer-S1-quantized-4bitTime series forecasting 13
- krishifer/sundial-base-128mTime series forecasting 11
- Feng800/sundial-base-128mTime series forecasting 10
How to load it
from datasets import load_dataset
ds = load_dataset("thuml/UTSD")