Timestral
Timestral’s two largest models rank first and second on every metric across three forecasting benchmarks, the authors report.
The family comprises four probabilistic time series forecasting models, ranging from 5 million to 209 million parameters. Its forecasting design separates prediction tokens from historical context and supports inputs whose future values are known. Training begins entirely on synthetic data, followed by a mix of real and synthetic data with longer context. Weights for all four models are available on Hugging Face, with inference code on GitHub.
ArXiv · AI/CL/LG's note
The family comprises four probabilistic time series forecasting models, ranging from 5 million to 209 million parameters. Its forecasting design separates prediction tokens from historical context and supports inputs whose future values are known. Training begins entirely on synthetic data, followed by a mix of real and synthetic data with longer context. Weights for all four models are available on Hugging Face, with inference code on GitHub.
ArXiv · AI/CL/LG's note
score 6