⬢github Python · 1.1K ★ +188 since we first saw it · pushed 19 d ago · Apache-2.0
mistralai/timestral
Repository for Timestral, the time-series foundation model by Mistral AI
Timestral is Mistral AI's family of time-series foundation models for multivariate probabilistic forecasting. This repo provides lightweight standalone inference (PyTorch, NumPy, safetensors) running on CPU or GPU. Four sizes (5M–209M) output 21 quantiles per step, support covariates, handle contexts up to 32,768 steps, and can be loaded from the Hugging Face Hub in a few lines of Python.
Why now: Freshly released by Mistral AI with claims of #1 rankings on GIFT-Eval, TIME, and fev-bench benchmarks, so it's trending on GitHub's most-starred new repos.
Who it is for: Data scientists and engineers who need forecasting (demand, metrics, sensors) without training custom models.
Stars over our 763 snapshots: 875 to 1.1K, since 8 d ago.
Where people talked about it
- ⬢github new repos, most starred 6 d ago
- ⬢github new repos, most starred 7 d ago
- ⬢github new repos, most starred 8 d ago
API: https://socialmediatrends-api.osmike.com/v1/repos/mistralai/timestral