Add OptimType.NONE in SplitTBE (defuse bwd and optim) #1820
Conversation
This file contains 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
✅ Deploy Preview for pytorch-fbgemm-docs canceled.
|
|
This pull request was exported from Phabricator. Differential Revision: D46618714 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jun 12, 2023
Summary: Pull Request resolved: pytorch#1820 This diff is the **frontend** part This diff introduces `OptimType.NONE`. Unlike other `OptimType`s, `OptimType.NONE` does not perform the optimizer step during SplitTBE's backward pass. With `OptimType.NONE`, SplitTBE deduplicates output gradients in the backward pass and generates a sparse gradient tensor (PyTorch's `sparse_coo_tensor`) for the device's weight (FQN: `weights_dev`). Currently, `OptimType.NONE` only supports the case where the embedding dimensions of all embedding tables are identical. Differential Revision: D46618714 fbshipit-source-id: 39309a9e7d03e43d8d8e81f1339091076557ed3e
|
This pull request was exported from Phabricator. Differential Revision: D46618714 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jun 12, 2023
Summary: Pull Request resolved: pytorch#1820 This diff is the **frontend** part This diff introduces `OptimType.NONE`. Unlike other `OptimType`s, `OptimType.NONE` does not perform the optimizer step during SplitTBE's backward pass. With `OptimType.NONE`, SplitTBE deduplicates output gradients in the backward pass and generates a sparse gradient tensor (PyTorch's `sparse_coo_tensor`) for the device's weight (FQN: `weights_dev`). Currently, `OptimType.NONE` only supports the case where the embedding dimensions of all embedding tables are identical. Differential Revision: D46618714 fbshipit-source-id: eac1334c2abea1561870a11ad1f1acbaa002e5c0
|
This pull request was exported from Phabricator. Differential Revision: D46618714 |
Differential Revision: D44392172 fbshipit-source-id: 60df0822cbc7d7341ab01ed7e0b67cb435a70395
Summary: Pull Request resolved: pytorch#1820 This diff is the **frontend** part This diff introduces `OptimType.NONE`. Unlike other `OptimType`s, `OptimType.NONE` does not perform the optimizer step during SplitTBE's backward pass. With `OptimType.NONE`, SplitTBE deduplicates output gradients in the backward pass and generates a sparse gradient tensor (PyTorch's `sparse_coo_tensor`) for the device's weight (FQN: `weights_dev`). Currently, `OptimType.NONE` only supports the case where the embedding dimensions of all embedding tables are identical. Differential Revision: D46618714 fbshipit-source-id: 3458c8b3e286573654b7a250f053f8fe8fd577cf
|
This pull request was exported from Phabricator. Differential Revision: D46618714 |
|
This pull request has been merged in 5cef9fc. |
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:
This diff is the frontend part
This diff introduces
OptimType.NONE. Unlike otherOptimTypes,OptimType.NONEdoes not perform the optimizer step during SplitTBE'sbackward pass. With
OptimType.NONE, SplitTBE deduplicates outputgradients in the backward pass and generates a sparse gradient tensor
(PyTorch's
sparse_coo_tensor) for the device's weight (FQN:weights_dev).Currently,
OptimType.NONEonly supports the case where the embeddingdimensions of all embedding tables are identical.
Differential Revision: D46618714