Datasets:

---

Real-TSF

/

TIME-Output

Like 1

Follow

Real-TSF 9

Evaluation results of Simons-297M on all 98 TIME tasks, for the leaderboard.

Simons is a family of open time series foundation models from Abel AI Lab (code, weights).

• Scores with scripts/compute_local_leaderboard.py, normalized by Seasonal Naive: MASE 0.637, CRPS 0.534.
• Protocol: multivariate series are forecast jointly (to_univariate=False, up to 32 variates per group); the context is the history before each window, up to 32,768 steps; nine quantiles; one forecast per window. Zero-shot: no TIME data in the training data.
• Code: experiments/simons_model.py and scripts/run_simons.sh in https://github.com/zqiao11/TIME/pull/18, run with simons at v0.1.0 on one GPU (bfloat16).
• Files: results/Simons-297M/<dataset>/<freq>/<term>/ with config.json, metrics.npz and predictions.npz for each task.

The other sizes: 25M, 60M, 178M, 297M.