Embedding grouper distinguish between prefetched vs non-prefetched table #1859
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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D55901328 |
…ble (pytorch#1859) Summary: If `prefetch_pipeline` is as fused parameter, training pipeline will try to call `prefetch()` in a separate stream one batch ahead of time. This process, is unfortunately consuming lots of extra memory. Practically it consumes 8~9x the size of input tensor at peak. Therefore, we wish to minimize the input tensor size to `prefetch()` call as much as possible. To achieve that, we don't want to mix tables that require prefetch and doesn't to be grouped to the same TBE. This diff will not change behavior for any jobs without cached embedding offloading. For embedding-offloaded jobs, this diff will slightly decrease the performance of TBE lookup as it result in more TBEs (and subsequently more kernels in forward and backward). but greatly increase the memory efficiency. Differential Revision: D55901328
levythu
force-pushed
the
export-D55901328
branch
from
f73b151 to
b4549e5
Compare
|
This pull request was exported from Phabricator. Differential Revision: D55901328 |
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Outdated suggestions cannot be applied.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:
If
prefetch_pipelineis as fused parameter, training pipeline will try to callprefetch()in a separate stream one batch ahead of time. This process, is unfortunately consuming lots of extra memory. Practically it consumes 8~9x the size of input tensor at peak.Therefore, we wish to minimize the input tensor size to
prefetch()call as much as possible. To achieve that, we don't want to mix tables that require prefetch and doesn't to be grouped to the same TBE.This diff will not change behavior for any jobs without cached embedding offloading.
For embedding-offloaded jobs, this diff will slightly decrease the performance of TBE lookup as it result in more TBEs (and subsequently more kernels in forward and backward). but greatly increase the memory efficiency.
Differential Revision: D55901328