NVIDIA / modulus-launch

Repo of optimized training recipes for accelerating PyTorch workflows of AI driven surrogates for physical systems
Apache License 2.0
56 stars 27 forks source link

🐛[BUG]: Inefficient inference in Vortex Shedding Example #60

Closed lucapegolotti closed 1 year ago

lucapegolotti commented 1 year ago

Version

0.1.0

On which installation method(s) does this occur?

No response

Describe the issue

In the Vortex Shedding example for MeshGraphNet, the rollout phase is performed once per component, i.e., two components for the velocity and one for the pressure (see here). However, each rollout simultaneously approximates all three components. Extracting all three components after one rollout could make the code more efficient.

Minimum reproducible example

No response

Relevant log output

No response

Environment details

No response