Earth-2 Model Intercomparison Project (MIP) is a python framework that enables climate researchers and scientists to inter-compare AI models for weather and climate.
I basically rewrote the implementation and based it more closely of graphcast's xarray interface rather than trying to remove it.
After 1 or two steps the output is nearly identical to the reference (I think up to machine precision of non-deterministic GPU ops)
Some new abstractions were helpful. I defined a functional TimeStepper abstraction that allows hiding model-specific state from the general purpose inference routines.
Checklist
[ ] I am familiar with the Contributing Guidelines.
[ ] New or existing tests cover these changes.
[ ] The documentation is up to date with these changes.
[ ] The CHANGELOG.md is up to date with these changes.
Earth-2 MIP Pull Request
Description
Resolves #130
I basically rewrote the implementation and based it more closely of graphcast's xarray interface rather than trying to remove it.
After 1 or two steps the output is nearly identical to the reference (I think up to machine precision of non-deterministic GPU ops)
Some new abstractions were helpful. I defined a functional
TimeStepper
abstraction that allows hiding model-specific state from the general purpose inference routines.Checklist
Dependencies