# Mistral AI recently released a 209M model that can predict future data locall... Canonical: https://social-archive.org/jo67/RrxXoqJItr Original URL: https://www.linkedin.com/posts/stephane-requena-0549b64_ai-gpu-share-7517568336906350592-7Iox/ Author: Stephane Requena Platform: linkedin Share mode: full ## Content Mistral AI recently released a 209M model that can predict future data locally on a CPU in ~140 milliseconds. Not every useful #AI model needs to talk (just like Jev) No #GPU required. It's a super small foundation model designed for time-series forecasting. Give it historical data like: ๐Ÿ“ˆ sales โšก power usage ๐Ÿญ sensor readings ๐Ÿ“ฆ demand/inventory ๐ŸŒก๏ธ telemetry โ€ฆand it predicts what comes next without training a new model for each dataset. The largest version stats ๐Ÿง  209M parameters ๐Ÿ’พ 837MB FP32 ๐Ÿ–ฅ๏ธ runs on CPU or GPU ๐Ÿ“š up to 32,768 historical time steps ๐Ÿ”ฎ forecasts up to 1,024 steps per forward pass ๐Ÿ“Š outputs 21 probability quantiles Mistral says it ranks #1 among pretrained/zero-shot models on all three major public forecasting benchmarks ๐Ÿฅ‡ GIFT-Eval ๐Ÿฅ‡ TIME ๐Ÿฅ‡ fev-bench It can get small like Timestral-5M โ†’ 21MB Mistral says on an 8-thread CPU: โšก 5M model โ†’ ~10 ms โšก 209M model โ†’ ~140 ms for a one-day forecast from three weeks of 15-minute data. All open source here : https://lnkd.in/eJhCTMfk ## Media 1. image: https://social-archiver-api.social-archive.org/media/archives/jo67/m4ECsmxAF1/media/0.jpg