Skip to content

Latest commit

 

History

History
26 lines (15 loc) · 2 KB

README.md

File metadata and controls

26 lines (15 loc) · 2 KB

Ai2 Climate Modeling Repostories

Our work is spread across a few key open-source, open-development repositories. These are roughly divided in to Applications and Libraries.

Applications

The domain specific language (DSL) and machine learning (ML) teams are each producing two top-level applications. DSL is creating a faster climate model, and ML is creating a better one:

  • (better) https://github.com/ai2cm/fv3net. A machine-learning capable climate model, machine learning training schemes, and diagnostics. This repository uses a monorepo style to suit the ML team's highly coordinated development approach.
  • (faster) https://github.com/ai2cm/pace. Pace is the Python version of the FV3GFS dynamical core and physical parameterizations based on the GT4Py domain-specific language (see Libraries below) which can run on x86 CPUs and NVIDIA GPUs.

We also maintain a fork of the baseline FV3GFS model with improved diagnostic capability and continuous integration.

Libraries

The applications above depend on a series of libraries:

We also maintain a fork of the baseline GT4Py library for our own rapid development before contributing new features upstream.

  • https://github.com/ai2cm/gt4py. GT4Py Python library for generating high-performance implementation of stencil kernels from a high-level defintion using regular Python functions.