Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/3483
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit e83342c with merge base c3d1680 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
Summary: Pull Request resolved: #3483 This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Differential Revision: D56401086
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
Summary: Pull Request resolved: #3483 This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Differential Revision: D56401086
Summary: This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Reviewed By: kimishpatel, mcr229 Differential Revision: D56401086
4ec6698 to
a414b5a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
Summary: This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Reviewed By: kimishpatel, mcr229 Differential Revision: D56401086
a414b5a to
7f2de9b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
Summary: This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Reviewed By: kimishpatel, mcr229 Differential Revision: D56401086
7f2de9b to
ae0853e
Compare
Summary: This diff introduces the to_edge_transform_and_lower API. The changes introduces are: - Adding support to the Parititioner class to register ops that it doesn't want to be composed - Changes to _program.py to add the implementation of to_edge_transform_and_lower() - Added a basic test case to test that Linear, SDPA & Linear + SDPA are not decomposed when asked and the corresponding backend consumes them. Reviewed By: kimishpatel, mcr229 Differential Revision: D56401086
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
ae0853e to
e83342c
Compare
|
This pull request was exported from Phabricator. Differential Revision: D56401086 |
|
This pull request has been merged in 2b91eba. |
Summary: ## Motivation Starting from iOS18, CoreML has added SDPA op, so there is no longer need to decompose torch SDPA ## Solution Following #3483, add `ops_not_to_decompose` in CoreML partitioner, then use `to_edge_transform_and_lower` API in llama export Pull Request resolved: #5258 Reviewed By: kirklandsign Differential Revision: D62550916 Pulled By: cccclai fbshipit-source-id: bc238a04ea9eb38341157cd2388b7cd17a506bf1
Summary:
This diff introduces the to_edge_transform_and_lower API. The changes introduces are:
Differential Revision: D56401086