Enable Optimizer Storing & Fix incomplete updates to Sharded EBC attributes in resharding #2911
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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D73530909 |
aporialiao
added a commit
to aporialiao/torchrec
that referenced
this pull request
Apr 23, 2025
…ibutes in resharding (pytorch#2911) Summary: Previously the dynamic sharding unit test was incomplete in truly verifying that a resharded EBC has all the attributes updated correctly. I ran into these issues when trying to enable optimizer state storing and DMP interface in D73049934 Main changes: 1. Add in dynamic sharding unit test's `are_sharded_ebc_modules_identical` the private attributes for ShardedEmbeddingCollection. This method will only compare primitive types or primitive reference types and tensors 1. This helped identify the gaps in current DS implementation - namely `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` not being updated correctly to reflect the new shard placements & order 2. Add in updates to `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` in reshard API for Sharded EBC. 3. Add in call to update Optimizer. The diff splits are not ideal, but the full optimizer unit test will be added in D73049934 Differential Revision: D73530909
aporialiao
force-pushed
the
export-D73530909
branch
from
7f8761d to
03cf9be
Compare
aporialiao
added a commit
to aporialiao/torchrec
that referenced
this pull request
Apr 23, 2025
…ibutes in resharding (pytorch#2911) Summary: Previously the dynamic sharding unit test was incomplete in truly verifying that a resharded EBC has all the attributes updated correctly. I ran into these issues when trying to enable optimizer state storing and DMP interface in D73049934 Main changes: 1. Add in dynamic sharding unit test's `are_sharded_ebc_modules_identical` the private attributes for ShardedEmbeddingCollection. This method will only compare primitive types or primitive reference types and tensors 1. This helped identify the gaps in current DS implementation - namely `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` not being updated correctly to reflect the new shard placements & order 2. Add in updates to `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` in reshard API for Sharded EBC. 3. Add in call to update Optimizer. The diff splits are not ideal, but the full optimizer unit test will be added in D73049934 Differential Revision: D73530909
aporialiao
force-pushed
the
export-D73530909
branch
from
03cf9be to
cf4f492
Compare
|
This pull request was exported from Phabricator. Differential Revision: D73530909 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D73530909 |
…ibutes in resharding (pytorch#2911) Summary: Pull Request resolved: pytorch#2911 Previously the dynamic sharding unit test was incomplete in truly verifying that a resharded EBC has all the attributes updated correctly. I ran into these issues when trying to enable optimizer state storing and DMP interface in D73049934 Main changes: 1. Add in dynamic sharding unit test's `are_sharded_ebc_modules_identical` the private attributes for ShardedEmbeddingCollection. This method will only compare primitive types or primitive reference types and tensors 1. This helped identify the gaps in current DS implementation - namely `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` not being updated correctly to reflect the new shard placements & order 2. Add in updates to `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` in reshard API for Sharded EBC. 3. Add in call to update Optimizer. The diff splits are not ideal, but the full optimizer unit test will be added in D73049934 Differential Revision: D73530909
|
This pull request was exported from Phabricator. Differential Revision: D73530909 |
aporialiao
force-pushed
the
export-D73530909
branch
from
cf4f492 to
cc3029f
Compare
aporialiao
added a commit
to aporialiao/torchrec
that referenced
this pull request
Apr 23, 2025
…ibutes in resharding (pytorch#2911) Summary: Previously the dynamic sharding unit test was incomplete in truly verifying that a resharded EBC has all the attributes updated correctly. I ran into these issues when trying to enable optimizer state storing and DMP interface in D73049934 Main changes: 1. Add in dynamic sharding unit test's `are_sharded_ebc_modules_identical` the private attributes for ShardedEmbeddingCollection. This method will only compare primitive types or primitive reference types and tensors 1. This helped identify the gaps in current DS implementation - namely `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` not being updated correctly to reflect the new shard placements & order 2. Add in updates to `module_sharding_plan`, `_embedding_dims`, `_uncombined_embedding_names`, `_uncombined_embedding_dims` in reshard API for Sharded EBC. 3. Add in call to update Optimizer. The diff splits are not ideal, but the full optimizer unit test will be added in D73049934 Differential Revision: D73530909
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.
Summary:
Previously the dynamic sharding unit test was incomplete in truly verifying that a resharded EBC has all the attributes updated correctly. I ran into these issues when trying to enable optimizer state storing and DMP interface in D73049934
Main changes:
Add in dynamic sharding unit test's
are_sharded_ebc_modules_identicalthe private attributes for ShardedEmbeddingCollection. This method will only compare primitive types or primitive reference types and tensors,i.e. Class types will be skipped.module_sharding_plan,_embedding_dims,_uncombined_embedding_names,_uncombined_embedding_dimsnot being updated correctly to reflect the new shard placements & orderAdd in updates to
module_sharding_plan,_embedding_dims,_uncombined_embedding_names,_uncombined_embedding_dimsin reshard API for Sharded EBC.Add in call to update Optimizer. The diff splits are not ideal, but the full optimizer unit test will be added in D73049934
Differential Revision: D73530909