fev-bench

Headlines · 3

1. 2d
   Mistral releases Timestral, a time-series foundation model family

   Mistral released Timestral, its first family of time-series foundation models, spanning 5M, 28M, 123M and 209M parameters with a shared architecture and training recipe. Mistral says its largest 209M-parameter model ranks first on the GIFT-Eval, TIME and fev-bench public benchmarks, beating larger models on key metrics, with results submitted to the GIFT-Eval and TIME leaderboards. The models produce probabilistic multivariate forecasts, accept past-only and known-future covariates and run without a GPU; Mistral cites roughly 10 ms for the 5M model and 140 ms for the 209M model forecasting one day ahead from three weeks of 15-minute history on an 8-thread server CPU. Code is available on GitHub under a modified MIT licence, with fine-tuning and domain adaptation offered through Mistral Forge and a technical report to follow.

   Mistral AI News · 🔥 4
2. 2d
   Databricks launches Predictive AI Functions in beta

   Databricks announced that Predictive AI Functions are now in beta, offering SQL-native functions that let analysts and business users run forecasting, driver analysis and anomaly detection directly in SQL, dashboards and Genie without building or managing models. The launch adds four functions — ai_top_drivers, ai_detect_anomalies, ai_predict_value and ai_predict_class — and improves ai_forecast, which is powered by a pretrained transformer-based time-series foundation model and needs no tuning, with support for holidays and external covariates. Databricks says ai_forecast outperforms a leading cloud data warehouse's equivalent and its own previous version on the fev-bench benchmark, though the comparison vendor is not named; within Genie only ai_top_drivers is available today, with the others listed as coming soon.

   Databricks Blog · 🔥 0
3. 2d
   TimeRSI: Self-improving agentic deployment for time-series foundation models

   The authors introduce TimeRSI, a self-improving agentic deployment system built around frozen time-series forecasting models: a coding agent translates task descriptions into executable pipeline changes, tests each revision on historical forecasting windows, and uses the observed errors to guide later revisions. At deployment time, an adaptive selector chooses among the resulting pipelines based on completed forecasts under similar conditions. The authors report that across 198 tasks on TIME and fev-bench, TimeRSI outperforms all submissions on six point and probabilistic forecasting metrics, and that controlled comparisons support feedback-guided revision over non-iterative pipeline search; code is available.

   Hugging Face · Papers · 🔥 0

Experience and discussion from the community

Share my fev-bench experienceAsk about fev-bench