Tasks and inputs
TimesFM 3.0 PyTorch is listed for Time Series Forecasting. It accepts time series and produces forecast. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Oct 7, 2026
TimesFM 3.0 is a numerical time-series forecasting checkpoint for PyTorch under a custom noncommercial licence.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
TimesFM 3.0 is a numerical time-series forecasting checkpoint for PyTorch under a custom noncommercial licence.
TimesFM takes time-series observations and returns forecasts. It is not a conversational model or a tool for generating prose about a chart. The model card describes a stacked mixing transformer with attention across variates and a patch-based forecasting interface. Architecture patch sizes should not be presented as a universal full-history or forecast limit.
Choose a series with a clear sampling interval and enough history to include its usual seasonality. Evaluate several rolling cutoffs, using only observations available before each cutoff. Compare against a seasonal baseline and inspect the error during spikes, missing-data periods and changes in trend. Keep the prediction horizon fixed when comparing two models.
Follow the publisher’s numerical preprocessing and PyTorch inference interface. Check the expected tensor shape, scaling, available covariates and returned forecast values before wiring the result into a dashboard or planning system. Distinguish uncertainty information supplied by the model from an interval calculated in your own application.
This checkpoint uses the TimesFM Non-Commercial License v1.0. An Apache licence on another TimesFM release does not establish commercial permission for this version. Keep hosted token pricing empty because the checkpoint’s forecasting interface does not establish language-model input and output charges. For operational use, measure the processing cost per series and the error on your own horizon.
Compare the model on the work you need to complete.
TimesFM 3.0 PyTorch is listed for Time Series Forecasting. It accepts time series and produces forecast. Use the publisher’s input format and preprocessing for your chosen deployment.
Evaluate forecasts at historical cutoffs using only the observations available at each cutoff. Compare with a simple baseline.
The weights are downloadable under TimesFM Non-Commercial License v1.0. Check runtime support and size the complete model and working context for your expected concurrency.