This repository was archived by the owner on Jan 6, 2023. It is now read-only.
imagenet example - add logic to broadcast most recent checkpoint from max_rank#93
Closed
kiukchung wants to merge 1 commit into
Closed
imagenet example - add logic to broadcast most recent checkpoint from max_rank#93kiukchung wants to merge 1 commit into
kiukchung wants to merge 1 commit into
Conversation
… max_rank Summary: Rationale for adding checkpoint broadcasting: - In our example we don't have access to globally visible storage - Each local rank =0 writes the checkpoint - when a container/node dies, the replacement container has no checkpoints (since it was lost with the node) - new nodes starts from scratch vs surviving nodes are ahead - the logic is to find the checkpoint with the max epoch and broadcast that Rationale for removing nnode==1 assertion for launcher with --with_etcd option. - you can run two agents on the same node to simulate a multi-node run - you can first start agent#1 by giving `--with_etcd` option - you can start agent#2 by copy pasting the rdzv info (from the logs) and passing the `--rdzv_id, --rdzv_backend, --rdzv_endpoint` from the first launch. Differential Revision: D20956704 fbshipit-source-id: 468d858db7fa9dfde7285270b2f0674faa52bcee
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D20956704 |
Contributor
|
This pull request has been merged in 1465dfc. |
fotstrt
pushed a commit
to eth-easl/elastic
that referenced
this pull request
Feb 17, 2022
… max_rank (pytorch#93) Summary: Pull Request resolved: pytorch#93 Rationale for adding checkpoint broadcasting: - In our example we don't have access to globally visible storage - Each local rank =0 writes the checkpoint - when a container/node dies, the replacement container has no checkpoints (since it was lost with the node) - new nodes starts from scratch vs surviving nodes are ahead - the logic is to find the checkpoint with the max epoch and broadcast that Rationale for removing nnode==1 assertion for launcher with --with_etcd option. - you can run two agents on the same node to simulate a multi-node run - you can first start agent#1 by giving `--with_etcd` option - you can start agent#2 by copy pasting the rdzv info (from the logs) and passing the `--rdzv_id, --rdzv_backend, --rdzv_endpoint` from the first launch. Reviewed By: tierex, drdarshan Differential Revision: D20956704 fbshipit-source-id: 3170e1bcbedf1a7522f3aeee23f0fc67cd038253
andrewjimpson9551
added a commit
to andrewjimpson9551/Elastic
that referenced
this pull request
Jul 29, 2025
… max_rank (#93) Summary: Pull Request resolved: pytorch/elastic#93 Rationale for adding checkpoint broadcasting: - In our example we don't have access to globally visible storage - Each local rank =0 writes the checkpoint - when a container/node dies, the replacement container has no checkpoints (since it was lost with the node) - new nodes starts from scratch vs surviving nodes are ahead - the logic is to find the checkpoint with the max epoch and broadcast that Rationale for removing nnode==1 assertion for launcher with --with_etcd option. - you can run two agents on the same node to simulate a multi-node run - you can first start agent#1 by giving `--with_etcd` option - you can start agent#2 by copy pasting the rdzv info (from the logs) and passing the `--rdzv_id, --rdzv_backend, --rdzv_endpoint` from the first launch. Reviewed By: tierex, drdarshan Differential Revision: D20956704 fbshipit-source-id: 3170e1bcbedf1a7522f3aeee23f0fc67cd038253
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
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
2 participants
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.
Summary:
Rationale for adding checkpoint broadcasting:
Rationale for removing nnode==1 assertion for launcher with --with_etcd option.
--with_etcdoption--rdzv_id, --rdzv_backend, --rdzv_endpointfrom the first launch.Differential Revision: D20956704