Model libraries and runtime SDKs
Use upstream Python libraries, ONNX Runtime Web, or an external provider client for the selected checkpoint.
Upstream model libraries
Chronos, TimesFM, and Moirai use different implementations and prediction interfaces. Follow each upstream repository, pin dependency and checkpoint versions, and verify supported devices. There is no current TSFM SDK that turns every checkpoint into a shared hosted inference contract.
Browser runtime
ONNX Runtime Web executes a compatible graph, but does not know the original model's preprocessing or output semantics. Keep shape validation and quantile interpretation with the model adapter. Run it in a worker, verify weights, and compare exported outputs with the upstream library.
Provider clients
Choose a provider client only after verifying its time-series model support and API contract. Do not assume that a generic text-generation client can execute a forecasting checkpoint. Store remote credentials outside the browser and test the actual error and timeout responses your application will receive.