Refactoring memory planning to allow running multiple algorithms #8440
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/8440
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit c2fa1c1 with merge base 94ec549 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
facebook-github-bot
pushed a commit
that referenced
this pull request
Feb 13, 2025
Summary: Pull Request resolved: #8440 Differential Revision: D69515056
facebook-github-bot
force-pushed
the
export-D69515056
branch
from
e6b5faf to
8eed6f3
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
facebook-github-bot
force-pushed
the
export-D69515056
branch
from
8eed6f3 to
af63ca9
Compare
facebook-github-bot
pushed a commit
that referenced
this pull request
Feb 13, 2025
Summary: Pull Request resolved: #8440 Differential Revision: D69515056
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
changed the title
tarun292
added a commit
that referenced
this pull request
Feb 27, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
af63ca9 to
32e3fa4
Compare
tarun292
requested review from
JacobSzwejbka,
SS-JIA and
larryliu0820
as code owners
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Feb 27, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Pull Request resolved: #8440 Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
32e3fa4 to
b323ab5
Compare
Contributor
|
What is the regression on lowering time? |
Contributor
Author
@JacobSzwejbka from a measurement on llama3 it takes about an additional 100ms to run 2 algorithms and pick the best one. |
tarun292
added a commit
that referenced
this pull request
Mar 7, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
b323ab5 to
ba0fa16
Compare
tarun292
added a commit
that referenced
this pull request
Mar 20, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Pull Request resolved: #8440 Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
69081ff to
27d74e2
Compare
facebook-github-bot
pushed a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
facebook-github-bot
force-pushed
the
export-D69515056
branch
from
27d74e2 to
a2cfdc9
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
facebook-github-bot
pushed a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
facebook-github-bot
force-pushed
the
export-D69515056
branch
from
a2cfdc9 to
a8802ae
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
a8802ae to
f365bba
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Pull Request resolved: #8440 Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
f365bba to
89496fc
Compare
tarun292
added a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
89496fc to
1e51dc0
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Mar 21, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Pull Request resolved: #8440 Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
1e51dc0 to
1a9b21f
Compare
tarun292
added a commit
that referenced
this pull request
Mar 22, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
1a9b21f to
ff1e53b
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
tarun292
added a commit
that referenced
this pull request
Mar 22, 2025
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Pull Request resolved: #8440 Reviewed By: JacobSzwejbka Differential Revision: D69515056
tarun292
force-pushed
the
export-D69515056
branch
from
ff1e53b to
f91623a
Compare
Summary: This diff introduces `memory_planning_algorithm_suite` which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed. The requirement for each of these algorithms is that they should generate a `MemoryAlgoResult` that contains the results of the memory planning done by that algorithm. These algos like before don't update the `TensorSpec` directly, but rather in `memory_planning_algorithm_suite` we figure out which algo gave us the best result and then update the `TensorSpec`'s with values (offsets etc.) returned by that algo. Reviewed By: JacobSzwejbka Differential Revision: D69515056
facebook-github-bot
force-pushed
the
export-D69515056
branch
from
f91623a to
c2fa1c1
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D69515056 |
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.
This diff introduces memory_planning_algorithm_suite which is a method that allows us to iterate through multiple memory planning algorithms and pick the one that gives us the best results i.e. least memory consumed.
The requirement for each of these algorithms is that they should generate a MemoryAlgoResult that contains the results of the memory planning done by that algorithm. These algos like before don't update the TensorSpec directly, but rather in memory_planning_algorithm_suite we figure out which algo gave us the best result and then update the TensorSpec's with values (offsets etc.) returned by that algo.
Differential Revision: D69515056