Report
Mistral releases 209M-parameter model for local time-series forecasting
A post says it forecasts up to 1,024 steps per pass without training a new model for each dataset.
TLDR
A post describes Mistral's 209M-parameter model as capable of forecasting from historical data on a CPU or GPU. It relays Mistral's claim that the model takes about 140 milliseconds on an eight-thread CPU to make a one-day forecast from three weeks of 15-minute data. Mistral also claims it ranks first among pretrained or zero-shot models on GIFT-Eval, TIME and fev-bench.
Combined views
—
1 Source, first seen ago
