Toy ECCO

Introduction

Toy version of ECCO estimation problem in Julia, and benchmarking of AD (Algorithmic Differentiation) tools including Enzyme.jl, Tapenade, Jax, ForwardDiff.jl, and Zygote.jl. The Optim.jl package is used for the optimization problem.

Collaborators

Shreyas Gaikwad, Max Trostel, Ellen Davenport, Gaël Forget

The problem

AD tools are hard to compare across different languages and algorithms because they all require slightly different implementation. This is an attempt to get some basic examples working with comparisons across languages (Python, Fortran, Julia) and open source AD tools (Enzyme, PyTorch, JAX, Tapenade).

Content

Summary slide

🚀🚀🚀 report/Toy ECCO Slide.pdf 🚀🚀🚀

notebooks and scripts folders

• bulkformulae_optimization.ipynb
• L96.ipynb
• L96_torch_jax.ipynb
• mountain_glacier.jl
• mountain_glacier_f90_model
• mountain_glacier_torch_jax.ipynb
• ForwardDiff_Zygote_Enzyme.ipynb
• optim_and_enzyme.ipynb

environment files

• environment.yml
• Project.toml
• Manifest.toml