Elastic Training support in DeepSpeed #2153
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…nd add ENABLE constants to init
…ct WORLD_SIZE computation
… training in docs, and added an assertion in runner for elastic training
tjruwase
reviewed
Jul 29, 2022
tjruwase
reviewed
Jul 29, 2022
jeffra
reviewed
Jul 29, 2022
jeffra
approved these changes
Jul 29, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds basic elastic training support to DeepSpeed using Elastic Agent in PyTorch.
Requirement:
PyTorch >= 1.11.x
Modifications:
Related Repos:
Megatron: https://github.com/microsoft/Megatron-DeepSpeed/tree/arpan/elastic_training
DeepSpeedExamples: https://github.com/microsoft/DeepSpeedExamples-internal/tree/arpan/elastic_scripts
Instructions to launch elastic training
Megatron Model: https://github.com/microsoft/Megatron-DeepSpeed/blob/arpan/elastic_training/examples/MoE/README_elastic.md
HelloDeepSpeed: https://github.com/microsoft/DeepSpeedExamples-internal/edit/arpan/elastic_scripts/elastic_training/README.md
Log:
We start training on allocation 1 (2 nodes) and drop one node after 40th iteration.
Workflow
PR is divided into two modules that can be enabled separately or simultaneously.
Runner module that focuses on relaunching of training in scale-up or scale-down event
Valid Batch Size computation in Elastic training
