torchft

This repository implements primitives and E2E solutions for doing a per-step fault tolerance so you can keep training if errors occur without interrupting the entire training job.

GETTING STARTED? See Install and Usage in the README.

Reference

• Process Groups
• Manager
• Optimizers
• Distributed Data Parallel
• LocalSGD
• Data
• Checkpointing
• Parameter Servers
• Coordination (Low Level API)

License

torchft is BSD 3-Clause licensed. See LICENSE for more details.

Copyright © Meta Platforms, Inc

• Terms of Use

• Privacy Policy