The PyTorch team introduces a framework for programming clusters

The PyTorch team from Meta, responsible for the open source machine learning framework PyTorch, introduced Monarch — a set of ready-made pieces of distributed programming programs. They make the work of coders easier.

Monarch, announced a few days ago by xrust, is a framework based on scalable messaging between actors. It allows users to program distributed systems in the same way as a single machine is programmed. This removes the complexity of distributed computing.

Monarch organizes processes, actors, and hosts into a scalable multi-dimensional array, or grid, that can be manipulated directly. Users can work with entire meshes or meshes using simple APIs. At the same time, Monarch automatically performs distribution and vectorization.

According to PyTorch experts, developers can write code as if nothing is happening. But when something goes wrong, Monarch quickly stops the entire program. Later, users can add granular error handling as needed, detecting failures and recovering from them.

Xrust The PyTorch team introduces a framework for programming clusters