Lightning is for building reactive, scalable, cost effective, easy-to-maintain and reliable ML products in the cloud without worrying about infrastructure. Lightning provides several engineering novelties to enable this:
#.**Reactivity**: Lightning allows you to run stateful components distributed across different machines, so you can design async, dynamic and reactive workflows in python, without having to define DAGs.
#.**Scalable & Cost-Effective**: Lightning provides a granular and simple way to run components preemptively or on-demand and on any desired resource such as CPU or GPU. It also enables you to easily transfer artifacts from one machine to another.
#.**Reliability**:
#.**Checkpointing**: Lightning apps can be paused and resumed from generated state and artifact-based checkpoints.
#.**Resilience**: Lightning has a strong fault-tolerance foundation. Your application can be written and tested to be resilient for cloud hazards at the component level.
#.**Testing Tools**: Lightning provides you with tools and best practices you can use to develop and test your application. All of our built-in templates have unit integration and end-to-end tests.
#.**Easy to maintain**:
#.**Easy Debugging**: Lightning apps can be debugged locally and in the cloud with **breakpoints** in any components.
#.**Non-Invasive**: Lightning is the glue that connects all parts of your workflow, but this is done in a non-invasive way by formalizing API contracts between components. In other words, your application can run someone else's code with little assumption.
