[dtensor][4/N] refactor dispatching logic and add propagator #90733
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
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
wanchaol
requested review from
H-Huang,
awgu,
kwen2501,
mrshenli,
pritamdamania87,
rohan-varma and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/90733
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 213b785: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Dec 13, 2022
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
wanchaol
changed the title
wanchaol
requested review from
XilunWu,
aazzolini and
fduwjj
and removed request for
pritamdamania87
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
XilunWu
approved these changes
Jan 18, 2023
Contributor
XilunWu
left a comment
XilunWu
left a comment
There was a problem hiding this comment.
Great work! Thx for refactoring op dispatching and sharding propagation facilities. some typos to fix.
| if op_call in _CURRENT_DECOMPOSITION_TABLE: | ||
| return _CURRENT_DECOMPOSITION_TABLE[op_call](*args, **kwargs) | ||
|
|
||
| # STEP 0. See if threre're user defined custom aten operator |
| # implementations. Custom operators take the highest priority | ||
| if custom_dispatch_ops is not None and str(op_call) in custom_dispatch_ops: | ||
| # dispatch to user defined custom distributed tensor ops | ||
| return custom_dispatch_ops[str(op_call)](*args, **kwargs) |
Contributor
There was a problem hiding this comment.
question: will this custom_dispatch_ops be deprecated once register_impl is no longer needed? I assume that eventually we want to get rid of register_impl and fully adopt propagation rules.
Collaborator
Author
There was a problem hiding this comment.
Yes we should deprecate this once we move all ops to use propagation rules
torch/distributed/_tensor/prop.py
Outdated
| if sharding_prop_func is None: | ||
| # step 1. If there's not even one sharding rule | ||
| # implemented for the operator, we fall back to | ||
| # local tensor compute, this is wront currently |
torch/distributed/_tensor/prop.py
Outdated
| # implemented for the operator, we fall back to | ||
| # local tensor compute, this is wront currently | ||
| # we will change the behavior to reshard to full | ||
| # replicate and do the computatation |
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This was referenced Jan 24, 2023
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. [ghstack-poisoned]
Collaborator
Author
|
@wanchaol has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. Differential Revision: [D42876251](https://our.internmc.facebook.com/intern/diff/D42876251) [ghstack-poisoned]
Collaborator
Author
|
@wanchaol has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. Differential Revision: [D42876251](https://our.internmc.facebook.com/intern/diff/D42876251) [ghstack-poisoned]
This PR refactors the dispatching logic to make it more clean, and isolate the sharding propagation logic out to a separate class. This is so that we can implement more complicated propagation features later. Differential Revision: [D42876251](https://our.internmc.facebook.com/intern/diff/D42876251) [ghstack-poisoned]
Contributor
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
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):
This PR refactors the dispatching logic to make it more clean, and
isolate the sharding propagation logic out to a separate class.
This is so that we can implement more complicated propagation features
later.
Differential Revision: D42876251