Spaces:
Running
Running
| title: TSFM Realworld Bench | |
| emoji: π | |
| colorFrom: blue | |
| colorTo: green | |
| sdk: gradio | |
| sdk_version: 4.44.0 | |
| app_file: app.py | |
| pinned: false | |
| license: apache-2.0 | |
| python_version: 3.11.9 | |
| # TSFM Realworld Bench Leaderboard | |
| Leaderboard for the **TSFM Realworld Benchmark** β evaluating time series foundation models on real-world univariate-target forecasting tasks. | |
| Results are submitted via pull request to the [TS-Live](https://github.com/zhouziyu02/TS-Live) repository. | |
| The leaderboard includes GIFT-Eval-style result slices aggregated by actual prediction | |
| length, domain, and frequency, normalized per configuration against Seasonal-Naive. | |
| The Overall tab additionally reports cumulative future-only MSE, RMSE, MAPE, quantile | |
| CRPS, and Real-Time Gain (RTG), together with temporal Stability, Kendall Improvement, | |
| Average Rank, pairwise Win Rate, and Elo. | |
| Public standings use canonical metric rows and trailing 24-hour, 7-day, and | |
| 30-day target-end windows. Forecast/target snapshots and local JSONL caches are | |
| kept out of the public Space repository. | |