[Autodiff] Deterministic gradient compute #7321
Merged
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
comaniac
reviewed
Jan 21, 2021
Comment on lines
353
to
358
| dag = tvm.auto_scheduler.ComputeDAG(grads) | ||
| repeat = 100 | ||
| for i in range(repeat): | ||
| grads = te.gradient(R, [X], head=ones) | ||
| new_dag = tvm.auto_scheduler.ComputeDAG(grads) | ||
| assert str(dag) == str(new_dag) |
There was a problem hiding this comment.
- Since auto_scheduler guarantees the DAG would be the same with the same given compute, you don't need to involve auto_scheduler in this test.
- I'm even not sure if we need this test because it seems cannot expose the real problem. IIUC, the non-deterministic behavior comes from the use of unordered_set, so you may still pass this pass when you're lucky even you break something. If that happens, this test becomes flaky. But I'd like to hear opinions from others. cc @yzhliu
There was a problem hiding this comment.
Since auto_scheduler guarantees the DAG would be the same with the same given compute, you don't need to involve auto_scheduler in this test.
Yes, I agree. I use auto_scheduler here only because it provides a hash_key for TE level ir. Any ideas about how to compare two Tensor?
I'm even not sure if we need this test because it seems cannot expose the real problem. IIUC, the non-deterministic behavior comes from the use of unordered_set, so you may still pass this pass when you're lucky even you break something. If that happens, this test becomes flaky. But I'd like to hear opinions from others.
Agree. I'm fine with removing the test.
comaniac
changed the title
yzhliu
approved these changes
Jan 27, 2021
alexwong
pushed a commit
to alexwong/tvm
that referenced
this pull request
Feb 11, 2021
* fix unstable compute * fix * fix * lint * sort linear equation * sort inequalities * fix * fix find * lint * fix find * lint
electriclilies
pushed a commit
to electriclilies/tvm
that referenced
this pull request
Feb 18, 2021
* fix unstable compute * fix * fix * lint * sort linear equation * sort inequalities * fix * fix find * lint * fix find * lint
Lokiiiiii
pushed a commit
to Lokiiiiii/tvm
that referenced
this pull request
Mar 2, 2021
* fix unstable compute * fix * fix * lint * sort linear equation * sort inequalities * fix * fix find * lint * fix find * lint
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Mar 2, 2021
* fix unstable compute * fix * fix * lint * sort linear equation * sort inequalities * fix * fix find * lint * fix find * lint
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.
te.gradientmay generate different (but equivalent) backward compute for a single forward, which may result in Ansor miss. This pr makes surete.gradientalways generates the same backward.cc @yzhliu @comaniac