[DataLoader] Switch to guaranteed determinism & add option to non_deterministic #53532
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
…erministic [ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit 01b6133 (more details on the Dr. CI page):
This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions to the (internal) Dr. CI Users group. |
… to non_deterministic"
Add a feature for the decorator `non_deterministic` with an argument of deterministic function to determine if the instance of this class is deterministic or not.
1. Non-argument decorator
```py
@non_deterministic
class ABC(IterDataPipe):
def __init__(self, datapipe, ...):
...
```
2. With function argument
e.g. GreedyJoin is deterministic when there is only one datapipe as input. But, it's non-deterministic with multiple datapipes
```py
@non_deterministic(lambda datapipes: len(datapipes) > 1)
class GreedyJoinIterDataPipe(IterDataPipe):
def __init__(self, datapipes, ...):
...
```
Test is updated [here](https://github.com/facebookexternal/torchdata/pull/15)
Differential Revision: [D26888825](https://our.internmc.facebook.com/intern/diff/D26888825)
[ghstack-poisoned]
Codecov Report
@@ Coverage Diff @@
## gh/ejguan/42/base #53532 +/- ##
=====================================================
- Coverage 77.34% 77.34% -0.01%
=====================================================
Files 1887 1887
Lines 184826 184845 +19
=====================================================
+ Hits 142958 142962 +4
- Misses 41868 41883 +15 |
VitalyFedyunin
approved these changes
Mar 14, 2021
… to non_deterministic"
Add a feature for the decorator `non_deterministic` with an argument of deterministic function to determine if the instance of this class is deterministic or not.
1. Non-argument decorator
```py
@non_deterministic
class ABC(IterDataPipe):
def __init__(self, datapipe, ...):
...
```
2. With function argument
e.g. GreedyJoin is deterministic when there is only one datapipe as input. But, it's non-deterministic with multiple datapipes
```py
@non_deterministic(lambda datapipes: len(datapipes) > 1)
class GreedyJoinIterDataPipe(IterDataPipe):
def __init__(self, datapipes, ...):
...
```
Test is updated [here](https://github.com/facebookexternal/torchdata/pull/15)
Differential Revision: [D26888825](https://our.internmc.facebook.com/intern/diff/D26888825)
[ghstack-poisoned]
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:
Add a feature for the decorator
non_deterministicwith an argument of deterministic function to determine if the instance of this class is deterministic or not.e.g. GreedyJoin is deterministic when there is only one datapipe as input. But, it's non-deterministic with multiple datapipes
Test is updated here
Differential Revision: D26888825