Mistral AI releases Timestral-5M, a 5M-parameter time series forecasting model
Overview
Mistral AI has released Timestral-5M, a 5-million-parameter zero-shot time series forecasting model, on Hugging Face.
It is the smallest checkpoint in a family that also includes 28M, 123M and 209M versions. The model uses a patch-based transformer with 8 layers, accepts up to 32,768 time steps of context, and forecasts up to 1,024 steps per pass.
Mistral reports a MASE score of 0.705 on GIFT-Eval and 0.652 on TIME. The checkpoint is released under a Modified MIT license.
Written by AI from the articles below · updated Oct 15, 12:07 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Mistral AI · new models on Hugging FaceOfficialMistral releases Timestral 5M time series forecasting model on Hugging Face
AIMistral AI has released the 5M-parameter Timestral checkpoint on Hugging Face, part of a zero-shot time series forecasting family that also includes 28M, 123M, and 209M versions. The model uses a patch-based transformer with 8 layers, accepts up to 32,768 time steps of context, and forecasts up to 1,024 steps per pass. It scores MASE 0.705 on GIFT-Eval and 0.652 on TIME, and is released under a Modified MIT license.
Heat trend
Not enough continuous observations to show a trend yet.