[CoreML Backend] Handle missing data types. #3134
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/3134
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 57fdb33 with merge base d3326a2 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
cymbalrush
force-pushed
the
cherry-pick-2984-by-pytorch_bot_bot_
branch
2 times, most recently
from
54bacc0 to
c66abec
Compare
|
Clang formatter is updated on the main branch but not on the release branch, this was causing the linter to fail. I updated the PR by adding this |
|
@cymbalrush Does the build-frameworks-ios failure sound relevant? |
@guangy10 failure is relevant, checking. |
|
@guangy10 failure is not related to the PR, checked with @shoumikhin . Could we go ahead with the merge? |
|
@huydhn any chance you know what the failing message mean? Feel like it's not related to this pr. It seems like it termniates during clean up stage. |
The failure is unrelated to this change and comes from this cherry pick https://hud.pytorch.org/pytorch/executorch/commit/e7e9e061426d32d0b8140b17fd9bca082d3a6755. To fix this, I think an additional cherry pick for #2996 and #3006 is needed. I have it here #3190, so let's see if it fixes the issue. |
cymbalrush
force-pushed
the
cherry-pick-2984-by-pytorch_bot_bot_
branch
from
c66abec to
f9b5015
Compare
|
There is a new failure in test-coreml-delegate. @cymbalrush, @cccclai |
Summary: **Changes** - The runtime was failing if it encountered a datatype not supported by CoreML framework. The changes add support for all the datatypes that are supported by coremltools basically if `CoreMLBackend` can export a model then runtime would execute it. Complex types are not supported because `coremltools` doesn't support it. - Improves and cleans the multiarray copying code. - Adds portable ops to CoreML executor so that it can run a partitioned model. **Testing** - Tested partitioned model `coreml_stories.pte` - Added multiarray copying tests. Pull Request resolved: #2984 Reviewed By: kirklandsign Differential Revision: D56003795 Pulled By: shoumikhin fbshipit-source-id: fa1c7846f9510d68c359aed6761aedb2d10c6f46 (cherry picked from commit d731866)
cymbalrush
force-pushed
the
cherry-pick-2984-by-pytorch_bot_bot_
branch
from
f9b5015 to
57fdb33
Compare
Changes
The runtime was failing if it encountered a datatype not supported by CoreML framework. The changes add support for all the datatypes that are supported by coremltools basically if
CoreMLBackendcan export a model then runtime would execute it. Complex types are not supported becausecoremltoolsdoesn't support it.Improves and cleans the multiarray copying code.
Adds portable ops to CoreML executor so that it can run a partitioned model.
Testing
coreml_stories.pte