Support DI sharding for FPE_EBC #2968
Closed
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: D74671655 |
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
May 27, 2025
Summary: Support models that have FeatureProcessedEmbeddingBagCollection. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding yet, so FPE_EBCs must be all sharded with the same (sharding_type, device). Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 3, 2025
Summary: Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 4, 2025
Summary: Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 4, 2025
Summary: Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 4, 2025
Summary: Pull Request resolved: pytorch#2968 Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Differential Revision: D74671655
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 6, 2025
Summary: Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
jingsh
suggested changes
Jun 9, 2025
| assert ( | ||
| shard_device_type == expected_device_type | ||
| ), f"Expected {expected_device_type} but got {shard_device_type} for FeatureProcessedEmbeddingBagCollection sharding device type" | ||
|
|
There was a problem hiding this comment.
since we assert here and the expected_device_type is always cuda, why don't we keep the signature of the method's env as a single ShardingEnv, and just pass the cuda's sharding evn to this?
There was a problem hiding this comment.
While right now FPE is used on cuda, we can support it on cuda/cpu in the future. Performing the check here allows us to de-duplicate the logic so we can ensure there's one sharding type for FPE for all sharding paths.
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
force-pushed
the
export-D74671655
branch
from
527d6cf to
14c9ff0
Compare
hannaxu
pushed a commit
to hannaxu/torchrec
that referenced
this pull request
Jun 16, 2025
Summary: Pull Request resolved: pytorch#2968 Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Reviewed By: faran928 Differential Revision: D74671655
Summary: Pull Request resolved: pytorch#2968 Support models that have FeatureProcessedEmbeddingBagCollection to be DI sharded. However, conservatively enforce that FPE itself can only be sharded on HBM and not across CPU as well. These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan. This doesn't support true hybrid sharding for FPE. Reviewed By: faran928 Differential Revision: D74671655
|
This pull request was exported from Phabricator. Differential Revision: D74671655 |
hannaxu
force-pushed
the
export-D74671655
branch
from
14c9ff0 to
2ec92ba
Compare
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:
Support models that have FeatureProcessedEmbeddingBagCollection.
These changes make sure we add QuantFeatureProcessedEmbeddingBagCollectionSharder as a recognized sharder, handle multiple envs needed for specifying DI sharding, and propagate TBE properly when processing the sharding plan.
Differential Revision: D74671655