Datasets:
Update evaluation results for ChakraTS
#64
by shubham-yhatlabs - opened
Updated evaluation results for ChakraTS on the TIME benchmark, from the current version of the model (updated November 2026): 98/98 tasks, config.json, metrics.npz and predictions.npz per task, plus a summary README. This replaces our earlier ChakraTS entry.
ChakraTS is a pretrained time-series model from YHat Labs, served through an API (model card: https://huggingface.co/yhatlabs/ChakraTS). The results use its pro preset, with the full history before each window sent as context and quantiles 0.1 to 0.9, written with save_window_predictions at the official task settings. scripts/compute_local_leaderboard.py on this folder gives MASE (norm.) 0.623 / CRPS (norm.) 0.522 (previous entry: 0.639 / 0.537).
Thank you for maintaining TIME.