Clockwork TorchPass
Zero-restart fault tolerance for distributed AI training
clockwork.io/workload-failover
Community-submitted · content unverified. This organization has not verified ownership or the information in this profile.
About
TorchPass keeps distributed AI training running through link flaps, GPU faults and node failures by migrating work off failing GPUs live instead of restarting from a checkpoint. It is software-only and works with TorchTitan, Megatron-LM, DeepSpeed, Slurm and Kubernetes. TorchPass Snapshots, launched October 2026, captures the state of a running multi-node job without training-code changes.
- Added via
- web
- Ownership
- unclaimed
Where to go
Product details
- Product kind
- Developer Tool
- Pricing model
- Enterprise