[subclasses] Use __slots__ for micro optim of flatten/unflatten #1211
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1211
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit f274127 with merge base 88d604f ( BROKEN TRUNK - The following job failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
IvanKobzarev
added a commit
that referenced
this pull request
Nov 1, 2024
ghstack-source-id: 78700dffdc8daf49e43a80e68e2a7888e681503c Pull Request resolved: #1211
facebook-github-bot
added
the
CLA Signed
…atten" Profiling the case from pytorch/pytorch#129457 found that using __slots__ a bit helps to reduce the cost of flatten (14us -> 11us). As a result 20 fp8 quantized weights flattening gets -40us (300us -> 260us). [ghstack-poisoned]
IvanKobzarev
added a commit
that referenced
this pull request
Nov 1, 2024
ghstack-source-id: 29e856540122dd6d0a8d3a522617234af70a6ca3 Pull Request resolved: #1211
weifengpy
reviewed
Nov 1, 2024
| @@ -258,6 +261,16 @@ def fsdp_post_all_gather( | |||
|
|
|||
|
|
|||
| class WeightWithDelayedFloat8CastTensor(torch.Tensor): | |||
|
|
|||
| __slots__ = [ | |||
There was a problem hiding this comment.
curious is __slots__ derived from __tensor_flatten__ ?
There was a problem hiding this comment.
Yes, here I just collected all attributes used in tensor_flatten, tensor_unflatten
…atten" Profiling the case from pytorch/pytorch#129457 found that using __slots__ a bit helps to reduce the cost of flatten (14us -> 11us). As a result 20 fp8 quantized weights flattening gets -40us (300us -> 260us). [ghstack-poisoned]
IvanKobzarev
added a commit
that referenced
this pull request
Nov 4, 2024
ghstack-source-id: 09af60f4537f21a491d21a02d99ad792185d32aa Pull Request resolved: #1211
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.
Stack from ghstack (oldest at bottom):
Profiling the case from pytorch/pytorch#129457 found that using slots a bit helps to reduce the cost of flatten (14us -> 11us). As a result 20 fp8 quantized weights flattening gets -40us (300us -> 260us).