forked from mjucker/MiMA
-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
An overarching issue to track progress towatds interfacing various ML models and benchmarking.
Oversees:
- TensorFlow GWP model for MiMA v2.0 #3
- PyTorch GWP model for MiMA v2.0 #4
- Bridging code to configure which GWP is being used #5
where more detail can be found.
@jatkinson1000 - covering #4 and #5
- Adapt
cg_drag.f90to accommodate drag forcing from various different approaches #11 - Create
.f90files with modules & subroutines for directly coupled ML methods #12 - Cannot pass multiple inputs to a model Cambridge-ICCS/FTorch#1
@SimonClifford - covering #3
- Construct a TensorFlow model to match the one in
src/shared/pytorch#13 - Create
.f90files with modules & subroutines for directly coupled ML methods #12
We then need to conduct benchmarking:
- Instrument MiMA for ReBench
- Plan a set of runs
- Perform the set of runs
Metadata
Metadata
Assignees
Labels
No labels