# Time Series Forecasting Datasets: AI Datasets Source: https://savrn.com/datasets/tasks/time-series-forecasting Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Datasets by Task # Time Series Forecasting Datasets 23 open-weight time series forecasting datasets in the SAVRN Model Hub, with Rogerdehe, Schraads and Astro Legacy Archive publishing the most. 23Datasets 19Publishers 7Licenses ## SAVRN's Take Nine datasets sit under this task, and two of the listed entries hold nearly all the bytes. GiftEvalPretrain from Salesforce AI Research bundles 71 univariate and 17 multivariate datasets across seven domains and 13 frequencies, 4.5 million series and 230 billion data points, with no leakage across the train and test split. At 397,938 downloads it is the entry for pretraining a forecasting model rather than tuning one. Open Climate Fix (Archives) publishes a subset of the Met Office UK Deterministic Dataset, converted from NetCDF into Zarr, our pick for solar output. The rest are narrow feeds: Polymarket CLOB WebSocket messages for microstructure work, IBM Quantum Heron calibration drift against space weather, and 14,612,276 rows of central bank exchange rates back to 1914. Disk comes first: GiftEvalPretrain is about 737 GB and the solar subset about 537 GB, so the two headline entries need over a terabyte of fast storage before training starts. The four smaller entries together fit under a gigabyte. On licenses, four of the nine are CC BY 4.0, one MIT, one Apache 2.0, and three are marked other, including the Polymarket archive; other is the category to open and read before a model trained on it earns revenue. We would test the no-leakage claim against our own held-out split before trusting it. SAVRN Research, 2026-09-18 ## Most Downloaded | Dataset | Publisher | License | Monthly downloads | | --- | --- | --- | --- | | [GiftEvalPretrain](https://savrn.com/datasets/giftevalpretrain) | Salesforce AI Research | apache-2.0 | 397.9k | | [met-office-uk-deterministic-solar](https://savrn.com/datasets/met-office-uk-deterministic-solar) | Open Climate Fix (Archives) | cc-by-4.0 | 177.8k | | [polymarket-l2-history](https://savrn.com/datasets/polymarket-l2-history) | Arthur | other | 46.7k | | [looked-up](https://savrn.com/datasets/looked-up) | Abdellah Sbai | cc0-1.0 | 23.1k | | [USDT-M_Perpetual_Futures](https://savrn.com/datasets/usdt-m-perpetual-futures) | Lin Xueyuan | mit | 17.7k | | [caliceo](https://savrn.com/datasets/caliceo) | Julien Chaumond | mit | 16.6k | | [klines-binance](https://savrn.com/datasets/klines-binance) | Rogerdehe | mit | 8.6k | | [qiskit-calibration-drift](https://savrn.com/datasets/qiskit-calibration-drift) | Charles Norton | cc-by-4.0 | 8.5k | | [klines-okx](https://savrn.com/datasets/klines-okx) | Rogerdehe | mit | 7.8k | | [klines-bybit](https://savrn.com/datasets/klines-bybit) | Rogerdehe | mit | 4.8k | ## Licenses | License | Datasets | Commercial use | | --- | --- | --- | | cc-by-4.0 | 9 | Yes | | other | 5 | Read the license | | mit | 5 | Yes | | not stated | 1 | Not stated | | cc0-1.0 | 1 | Read the license | | cc-by-nc-sa-4.0 | 1 | Not without separate permission | | apache-2.0 | 1 | Yes | ## Who Publishes Them | Publisher | Datasets | | --- | --- | | [Rogerdehe](https://savrn.com/model-publishers/rogerdehe) | 3 | | [Schraads](https://savrn.com/model-publishers/schraads) | 2 | | [Astro Legacy Archive](https://savrn.com/model-publishers/astro-legacy-archive) | 2 | | [Tharindu Perera](https://savrn.com/model-publishers/tharinduperera) | 1 | | [Subatomic Physics & Detector Telemetry Lab (SPDTL)](https://savrn.com/model-publishers/subatomic-telemetry-lab) | 1 | | [Abdellah Sbai](https://savrn.com/model-publishers/sbaiiiiii) | 1 | ## All 23 Datasets Dataset · Time series forecasting ### [GiftEvalPretrain](https://savrn.com/datasets/giftevalpretrain) [Salesforce AI Research](https://savrn.com/model-publishers/salesforce) Pretraining dataset aligned with GIFT-Eval that has 71 univariate and 17 multivariate datasets, spanning seven domains and 13 frequencies, totaling 4.5 million time series and 230 billion data points. Notably this collection of data has no leakage issue with the train/test split and can be used to pretrain foundation models that can be fairly evaluated on GIFT-Eval. This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model. We encourage… Publicly accessible apache-2.0 1M