[FSDP][Perf] Pre-allocate sharded grad in default stream #90617
Closed
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
[ghstack-poisoned]
awgu
requested review from
mrshenli,
zhaojuanmao,
pritamdamania87,
rohan-varma,
H-Huang,
kwen2501 and
wanchaol
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/90617
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Failures, 10 PendingAs of commit 6007ced: The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
release notes: distributed (fsdp)
This was referenced Dec 10, 2022
[FSDP][Perf] Pre-allocate sharded grad in default stream; save one copy when downcasting grad
#90616
Closed
[ghstack-poisoned]
[ghstack-poisoned]
Every sharded strategy always allocates a (padded) sharded gradient as the target tensor for the reduce-scatter. This PR moves that allocation to the default stream instead of the post-backward stream. Minor: This PR changes from `state.device` to `handle.device`, which has no semantic difference. It is just better to lower as much logic as possible to the `handle` when appropriate. [ghstack-poisoned]
Every sharded strategy always allocates a (padded) sharded gradient as the target tensor for the reduce-scatter. This PR moves that allocation to the default stream instead of the post-backward stream. Minor: This PR changes from `state.device` to `handle.device`, which has no semantic difference. It is just better to lower as much logic as possible to the `handle` when appropriate. [ghstack-poisoned]
Every sharded strategy always allocates a (padded) sharded gradient as the target tensor for the reduce-scatter. This PR moves that allocation to the default stream instead of the post-backward stream. Minor: This PR changes from `state.device` to `handle.device`, which has no semantic difference. It is just better to lower as much logic as possible to the `handle` when appropriate. [ghstack-poisoned]
Every sharded strategy always allocates a (padded) sharded gradient as the target tensor for the reduce-scatter. This PR moves that allocation to the default stream instead of the post-backward stream. Minor: This PR changes from `state.device` to `handle.device`, which has no semantic difference. It is just better to lower as much logic as possible to the `handle` when appropriate. [ghstack-poisoned]
|
This PR got absorbed into the previous PR by accident when dealing with rebase conflicts. |
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:
HandleConfigfor mixed precision #90631 [FSDP] SanitizeHandleConfigfor mixed precisionreduce_dtype#90615 [FSDP] Tighten post-bwd cast toreduce_dtype_storage()in test file #90622 [FSDP][Easy] Move to_storage()in test file_stream_to_namefor debugging #90611 [FSDP] Save_stream_to_namefor debuggingDTensor,use_orig_params=True#90562 [Reland][FSDP] Another fix forDTensor,use_orig_params=TrueEvery sharded strategy always allocates a (padded) sharded gradient as the target tensor for the reduce-scatter. This PR moves that allocation to the default stream instead of the post-backward stream.
Minor: This PR changes from
state.devicetohandle.device, which has no semantic difference. It is just better to lower as much logic as possible to thehandlewhen appropriate.