Welcome to the TorchSnapshot documentation!

TorchSnapshot is a PyTorch library for adding fault tolerance to large-scale PyTorch distributed training workloads.

Installation instructions

TorchSnapshot API

Contents:

• Getting Started
  • Describing the Application State
  • Taking a Snapshot
  • Restoring From a Snapshot
  • Distributed Snapshot
  • Snapshot Content Access
  • Taking a Snapshot Asynchronously
  • Reproducibility
  • Elasticity (Experimental)
• API Reference

Examples

• Simple example

• Using TorchSnapshot with DistributedDataParallel (DDP)

• Using TorchSnapshot with TorchRec