Apply default precision from bench models to XLA #6133
Merged
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 introduces a change fixing a disparity between instructions' precision of XLA, and TorchInductor. For example: After calling torch.half() or any other buffer/parameter casting TorchInductor will cast all non-fp16 floating types to fp16, and program input shape will correctly be reflected by this. OTOH XLA needs additional environment variables to perform casting upon lowering. This PR reads the default precision set by TorchBench models for inference, and training and sets the appropriate env vars.
golechwierowicz
force-pushed
the
olechwierowicz/apply-default-torchbench-precision
branch
from
36e7cf3 to
42d44eb
Compare
cota
approved these changes
Dec 13, 2023
Collaborator
|
Thanks Greg! Please land this and I'll run another nightly with it today. |
frgossen
reviewed
Dec 13, 2023
| del benchmark | ||
| gc.collect() | ||
|
|
||
| def apply_default_precision_config(self, test, benchmark): |
Collaborator
There was a problem hiding this comment.
Thanks for finding this and adding it to the benchmarks.
I am wondering if this is the right place to fix this.
Is this something we can add in the dynamo bridge when lowering to XLA?
Collaborator
There was a problem hiding this comment.
This is a workaround. See b/316126405
Collaborator
There was a problem hiding this comment.
I wonder if we should add these things to benchmarking or if that is just more confusing overall. We should at least add a comment in these cases.
chunnienc
pushed a commit
to chunnienc/xla
that referenced
this pull request
Dec 14, 2023
golechwierowicz
added a commit
that referenced
this pull request
Jan 12, 2024
bhavya01
pushed a commit
that referenced
this pull request
Apr 22, 2024
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.
This PR introduces a change fixing a disparity between instructions' precision of XLA, and TorchInductor.
For example:
After calling
torch.half()or any other buffer/parameter casting TorchInductor will cast all non-fp16 floating types tofp16, and program input shape will correctly be reflected by this. OTOH XLA needs additional environment variables to perform casting upon lowering.This PR reads the default precision set by TorchBench models for inference, and training and sets the appropriate env vars.