GIFT-Eval
Evaluate forecasting models across varied time scales, domains and prediction horizons.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The original paper describes 23 datasets across seven domains; current submission instructions specify 97 dataset configurations. Authors: Taha Aksu, Gerald Woo, Juncheng Liu, Xu Liu, Chenghao Liu, Silvio Savarese, Caiming Xiong and Doyen Sahoo.
Methodology
Use the supplied chronological splits and report point and probabilistic forecast errors. Current zero-shot eligibility excludes both test leakage and training on any GIFT-Eval training split; pretrained, fine-tuned and agentic submissions are distinct. Scoring corrections and eligibility rules have changed since the original paper. Preserve the data snapshot and submission category. GiftEvalPretrain is a separate training resource, and GIFT-Eval-CTX is a distinct followup.