How long was the training time roughly?

Rose-STL-Lab / dyffusion

[NeurIPS 2023] A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting

https://salvarc.github.io/blog/2023/dyffusion

Apache License 2.0

131 stars 12 forks source link

How long was the training time roughly? #2

Closed shiyu5feng closed 6 months ago

shiyu5feng commented 6 months ago

Hello, very great work. I am very interested in your work. May I ask on what GPU your experiments are trained? And how long was the training time roughly?

salvaRC commented 6 months ago

Thank you! For reference, on an A100 GPU the Navier-Stokes training for DYffusion takes ~13h for interpolator net + 25h for the forecaster net. Some experiments were also run an a single GeForce RTX 3090, where the forecaster net alone can take ~35h to train. However, for these datasets you should be able to get good performance even if you train for fewer epochs.

shiyu5feng commented 6 months ago

OK，Thank you.