January 2022 Summaries

AutoDevTech, a platform that uses machine learning to analyze code coverage, churn, and engineering engagement, adopted Grid to scale its model training without building a complex internal MLOps environment. After relying on traditional regression models and evaluating alternatives such as Horovod and SageMaker, the company chose Grid and PyTorch Lightning to support large-scale distributed training with minimal code changes. Operating Grid within its existing AWS virtual private cloud allowed the team to retain its infrastructure while using features including datastore management, resource visibility, Spot instances, and planned auto-resume capabilities. AutoDevTech reports that the platform enables engineers to run larger experiments, evaluate model design decisions more quickly, and focus more of their effort on machine learning development.