-
Notifications
You must be signed in to change notification settings - Fork 807
Arm backend: Initial bfloat16 support. #16740
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
Erik-Lundell
wants to merge
8
commits into
pytorch:main
Choose a base branch
from
Erik-Lundell:upstream-bf16
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
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
There is no reason for TosaArg to have a tosa_spec attribute, it is available in self.tosa_spec for node visitors. Remove unused tosa_spec argument in map_dtype to unify __validate() approach. Signed-off-by: Erik Lundell <erik.lundell@arm.com> Change-Id: I0c13964af42bd579709d8115f9777d484f118566
Update add and transpose node visitors as examples. Since TosaArg validates that bf16 tensors are only created for EXT-BF16, no need to check again in the node visitor. Though transpose support is extended to support rank-4 adds, tests and proper support will be done in a follow-up patch. Signed-off-by: Erik Lundell <erik.lundell@arm.com> Change-Id: I48adb9d106293df7aea415fcfe25b652202bab28
abs, average_pool_2d, cat, ceil and clamp mapping to TOSA operators ABS, ARGMAX, AVG_POOL2D, CAT, CEIL, CLAMP Required refactoring some serialization going through numpy since numpy doesn't have native support of bfloat16. Where possible, avoid dtype handling all together by viewing as uint8. In process node, make a special case for bf16 and use ml_dtypes. Signed-off-by: Erik Lundell <erik.lundell@arm.com> Change-Id: I92c8cb204396d10d0809db93822dea7fe71ec8a0
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16740
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Cancelled JobAs of commit cd898de with merge base 59aa69d ( CANCELLED JOB - The following job was cancelled. Please retry:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Collaborator
|
Rebasing to verify that the cortex-m fails goes away |
Signed-off-by: Erik Lundell <erik.lundell@arm.com> Change-Id: I6c95f2c782913f8ea0a928102043547f0b87f5c4
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Labels
ciflow/trunk
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
partner: arm
For backend delegation, kernels, demo, etc. from the 3rd-party partner, Arm
release notes: arm
Changes to the ARM backend delegate
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.
Add the necessary infrastructure and add support for some initial ops. Currently, only the TOSA backend supports the dtype.
cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai