pytorch
RFC-0022: Proposal for adding new restart policy to torch.distributed…
First, moving that direction is a great step! thanks!
Second, I would highly suggest looking into Elastic Horovod, I think they managed to get few things right that would be big additions here:
- Graceful node removal = continuing as-is without losing a single batch!
- Providing very helpful classes for handling full state syncing (including weights, sampler, optimizer, primitive types, etc..), this is vital for simple true graceful node removal
- The idea of doing memory (not disk) checkpointing of model weights seems interesting (much cheaper), esp. given this should be a fault-tolerant distributed system
Also, would this RFC support dynamically adding new nodes while training ?
Hi @aivanou!
Thank you for your pull request.
We require contributors to sign our Contributor License Agreement, and yours needs attention.
You currently have a record in our system, but the CLA is no longer valid, and will need to be resubmitted.
Process
In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.
Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.
If you have received this in error or have any questions, please contact us at [email protected]. Thanks!
