BUTTER: An Empirical Deep Learning Experimental Framework

NREL's BUTTER Empirical Deep Learning Framework enables researchers to run high volumes of computational experiments, including machine learning experiments, in a highly distributed, asynchronous way.

The framework was designed for asynchronous, unpredictable, and occasionally unreliable worker jobs to execute on any number of computing systems, including laptops, servers, cloud resources, clusters, and high-performance supercomputers.

The BUTTER deep learning dataset is built from the empirical results of millions of computational experiments run across multiple high-performance computing systems. It helps inform the impact of various parameters and settings on the performance and efficiency of deep learning training.


The BUTTER dataset is available via the U.S. Department of Energy Office of Scientific and Technical Information.

Software Framework

The application can be accessed from a GitHub repository.


Charles Tripp

Researcher IV, Computational Science