Experiments for Einsum Benchmark Dataset

The full dataset is stored at zenodo. You can conveniently access it using our wrapper package.

Experiments

This repository contains four different experiments. The explaination and code for each experiment is contained in one sub folder:

• Experiment 1 - Scalability of processing large contraction paths
• Experiment 2 - Asymmetric tensor sizes in pairwise contractions
• Experiment 3 - Sparse intermediate tensors
• Experiment 4 - Data types

Prerequisites

The experiments were executed on the follwing environment:

• python (3.10.9)
• numpy (1.26.4)
• opt_einsum (3.3.0)
• cotengra (0.6.2)
• kahypar (1.3.5)
• tensorflow (2.16.1)
• torch (1.12.1)
• jax (0.4.28)
• jaxlib (0.4.28)
• pyinstrument (4.6.2)
• tqdm (4.66.4)
• sqlite (3.45.3)

You can use the env.yml file to create the conda environment einsumbm using

conda env create -f env.yml

To remove the environment first deactivate it

conda deactivate

and then remove it

conda env remove -n einsumbm