Best practice for deploying TimesFM on SageMaker

google-research / timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

https://research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting/

Apache License 2.0

3.81k stars 324 forks source link

Best practice for deploying TimesFM on SageMaker #142

Open safaricd opened 2 months ago

safaricd commented 2 months ago

👋 What are the best practices for running TimesFM on Amazon SageMaker, considering the dynamic nature of horizon_len. Can the model be loaded and initialized first, so I can specify the horizon_len parameter at inference/forecast time?

rajatsen91 commented 2 months ago

I think its best to set the horizon_len to the maximum of what you would need at init. This way it will be jitted and all subsequent forecasts would be very fast.