Recompute linear_cache_indices for pipeline prefetching #2147
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
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
✅ Deploy Preview for pytorch-fbgemm-docs canceled.
|
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Nov 21, 2023
Summary: When pipeline prefetching is enabled (`prefetch_pipeline=True`) for `EmbeddingLocation.MANAGED_CACHING`, TBE has to update `lxu_cache_locations` to ensure cache consistency before the backward pass. The `lxu_cache_locations` update requires `linear_cache_indices` as an input. Prior to this diff, TBE keeps `linear_cache_indices` alive after prefetching until the tensor is used for the `lxu_cache_locations` update. This puts a lot of pressure to the memory space requirement limiting the enablement of pipeline prefetching for some models. This diff addresses the memory limitation issue by recomputing `linear_cache_indices` when it is needed. Differential Revision: D50983176
sryap
force-pushed
the
export-D50983176
branch
from
e93e108 to
26da350
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
Summary: When pipeline prefetching is enabled (`prefetch_pipeline=True`) for `EmbeddingLocation.MANAGED_CACHING`, TBE has to update `lxu_cache_locations` to ensure cache consistency before the backward pass. The `lxu_cache_locations` update requires `linear_cache_indices` as an input. Prior to this diff, TBE keeps `linear_cache_indices` alive after prefetching until the tensor is used for the `lxu_cache_locations` update. This puts a lot of pressure to the memory space requirement limiting the enablement of pipeline prefetching for some models. This diff addresses the memory limitation issue by recomputing `linear_cache_indices` when it is needed. Reviewed By: jspark1105 Differential Revision: D50983176
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Nov 21, 2023
Summary: When pipeline prefetching is enabled (`prefetch_pipeline=True`) for `EmbeddingLocation.MANAGED_CACHING`, TBE has to update `lxu_cache_locations` to ensure cache consistency before the backward pass. The `lxu_cache_locations` update requires `linear_cache_indices` as an input. Prior to this diff, TBE keeps `linear_cache_indices` alive after prefetching until the tensor is used for the `lxu_cache_locations` update. This puts a lot of pressure to the memory space requirement limiting the enablement of pipeline prefetching for some models. This diff addresses the memory limitation issue by recomputing `linear_cache_indices` when it is needed. Reviewed By: jspark1105 Differential Revision: D50983176
sryap
force-pushed
the
export-D50983176
branch
from
26da350 to
6fbcef2
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Nov 21, 2023
Summary: When pipeline prefetching is enabled (`prefetch_pipeline=True`) for `EmbeddingLocation.MANAGED_CACHING`, TBE has to update `lxu_cache_locations` to ensure cache consistency before the backward pass. The `lxu_cache_locations` update requires `linear_cache_indices` as an input. Prior to this diff, TBE keeps `linear_cache_indices` alive after prefetching until the tensor is used for the `lxu_cache_locations` update. This puts a lot of pressure to the memory space requirement limiting the enablement of pipeline prefetching for some models. This diff addresses the memory limitation issue by recomputing `linear_cache_indices` when it is needed. Reviewed By: jspark1105 Differential Revision: D50983176
sryap
force-pushed
the
export-D50983176
branch
from
6fbcef2 to
9ce5068
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Nov 21, 2023
Summary: When pipeline prefetching is enabled (`prefetch_pipeline=True`) for `EmbeddingLocation.MANAGED_CACHING`, TBE has to update `lxu_cache_locations` to ensure cache consistency before the backward pass. The `lxu_cache_locations` update requires `linear_cache_indices` as an input. Prior to this diff, TBE keeps `linear_cache_indices` alive after prefetching until the tensor is used for the `lxu_cache_locations` update. This puts a lot of pressure to the memory space requirement limiting the enablement of pipeline prefetching for some models. This diff addresses the memory limitation issue by recomputing `linear_cache_indices` when it is needed. Reviewed By: jspark1105 Differential Revision: D50983176
sryap
force-pushed
the
export-D50983176
branch
from
9ce5068 to
693bb6f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
sryap
force-pushed
the
export-D50983176
branch
from
693bb6f to
6c8e0a6
Compare
|
This pull request was exported from Phabricator. Differential Revision: D50983176 |
|
This pull request has been merged in 37111f5. |
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:
When pipeline prefetching is enabled (
prefetch_pipeline=True) forEmbeddingLocation.MANAGED_CACHING, TBE has to updatelxu_cache_locationsto ensure cache consistency before the backwardpass. The
lxu_cache_locationsupdate requireslinear_cache_indicesas an input. Prior to this diff, TBE keepslinear_cache_indicesalive after prefetching until the tensor isused for the
lxu_cache_locationsupdate. This puts a lot ofpressure to the memory space requirement limiting the enablement of
pipeline prefetching for some models. This diff addresses the memory
limitation issue by recomputing
linear_cache_indiceswhen it isneeded.
Differential Revision: D50983176