Skip to content

schuups/DSM500_FPR

Repository files navigation

DSM500 regarding NVIDIA Modulus's GraphCast PyTorch re-implementation

This repository contains a streamlined version of the code, scripts and results from the DSM500 project execution, providing clear, concise insights and reproducibility information - that matches and supports the Final Project Report reading flow. Given the refactoring, it might be that in the movement of files from the old repository some file paths might have not been updated. This might be the case in particular for notebooks. Nonetheless, the logic is still valid.

Key folders

Folder Contents description
fourcastnet-92260c1 The FourCastNet codebase used to train the external reference model (commit 92260c1)
modulus-a5275d8 The codebase copied from the official NVIDIA Modulus github repository (commit a5275d8)
modulus-baseline The refactored version of GraphCast used as a baseline for performance improvements.
modulus-dry-run The version of Modulus GraphCast used to collect the dry-run results
modulus-improved The version of Modulus GraphCast on which the improvements have been implemented.
supplementary_files Contains supporting code and scripts, e.g. used to generate dataset means and std values needed for normalisation, or to perform inference on a trained model.

About

Analysis and benchmarking of the GraphCast PyTorch reimplementation in the NVIDIA PhysicsNeMo library

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors