[ExecuTorch][Export][1/N] Export API pipeline re-architecture, making it composable #12936
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
… it composable RFC: #12660 This diff introduces composable architecture for the export pipeline. Changes: 1. Introduces notion of `PipelineArtifact` which encompasses stage artifacts and run context necessary to execute a stage and the pipeline. 1. Move all stages to new file to better architect the component 1. Export api will now accept `pipeline_stages` and honors the sequence if provided otherwise fallback to default sequence: `source_transform -> quantize -> torch.export -> to_edge_transform_and_lower -> to_executorch` 1. Validate if pipeline provided is valid, error out if not. 1. Add unittests to test pipeline sequence and stages With this one can execute a partial pipeline, for example, one can just do `TORCH_EXPORT -> TO_EDGE_TRANSFORM_AND_LOWER -> TO_EXECUTORCH` Current limitation: 1. `TORCH_EXPORT` stage is mandatory to avoid someone passing incorrect input such as `ExportedProgram` instead of nn.module, this enforcement will anchor the pipeline to have expected stages to run. - This limitation will be removed if we choose to add the support to take `ExportedProgram` as input. Note: 1. this diff is large because I moved stages to separate file and added bunch of unittests but core functionality that is added can be reviewed in `export.py` file. CC: @digantdesai 2. Export component was supposed to mimic stages design in tester and now it is. Fixes: #12928 Differential Revision: [D79120574](https://our.internmc.facebook.com/intern/diff/D79120574/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12936
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 17 PendingAs of commit 3407bd6 with merge base f497f7f ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
This was referenced Jul 28, 2025
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D79120574 |
This PR needs a
|
…ure, making it composable" RFC: #12660 This diff introduces composable architecture for the export pipeline. Changes: 1. Introduces notion of `PipelineArtifact` which encompasses stage artifacts and run context necessary to execute a stage and the pipeline. 1. Move all stages to new file to better architect the component 1. Export api will now accept `pipeline_stages` and honors the sequence if provided otherwise fallback to default sequence: `source_transform -> quantize -> torch.export -> to_edge_transform_and_lower -> to_executorch` 1. Validate if pipeline provided is valid, error out if not. 1. Add unittests to test pipeline sequence and stages With this one can execute a partial pipeline, for example, one can just do `TORCH_EXPORT -> TO_EDGE_TRANSFORM_AND_LOWER -> TO_EXECUTORCH` Current limitation: 1. `TORCH_EXPORT` stage is mandatory to avoid someone passing incorrect input such as `ExportedProgram` instead of nn.module, this enforcement will anchor the pipeline to have expected stages to run. - This limitation will be removed if we choose to add the support to take `ExportedProgram` as input. Note: 1. this diff is large because I moved stages to separate file and added bunch of unittests but core functionality that is added can be reviewed in `export.py` file. CC: digantdesai 2. Export component was supposed to mimic stages design in tester and now it is. Fixes: #12928 Differential Revision: [D79120574](https://our.internmc.facebook.com/intern/diff/D79120574/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D79120574 |
digantdesai
approved these changes
Jul 31, 2025
Contributor
digantdesai
left a comment
digantdesai
left a comment
There was a problem hiding this comment.
LGTM. Thanks for refactoring. No functional changes right?
larryliu0820
approved these changes
Jul 31, 2025
Contributor
larryliu0820
left a comment
larryliu0820
left a comment
There was a problem hiding this comment.
Reviewed internally
…ure, making it composable" RFC: #12660 This diff introduces composable architecture for the export pipeline. Changes: 1. Introduces notion of `PipelineArtifact` which encompasses stage artifacts and run context necessary to execute a stage and the pipeline. 1. Move all stages to new file to better architect the component 1. Export api will now accept `pipeline_stages` and honors the sequence if provided otherwise fallback to default sequence: `source_transform -> quantize -> torch.export -> to_edge_transform_and_lower -> to_executorch` 1. Validate if pipeline provided is valid, error out if not. 1. Add unittests to test pipeline sequence and stages With this one can execute a partial pipeline, for example, one can just do `TORCH_EXPORT -> TO_EDGE_TRANSFORM_AND_LOWER -> TO_EXECUTORCH` Current limitation: 1. `TORCH_EXPORT` stage is mandatory to avoid someone passing incorrect input such as `ExportedProgram` instead of nn.module, this enforcement will anchor the pipeline to have expected stages to run. - This limitation will be removed if we choose to add the support to take `ExportedProgram` as input. Note: 1. this diff is large because I moved stages to separate file and added bunch of unittests but core functionality that is added can be reviewed in `export.py` file. CC: digantdesai 2. Export component was supposed to mimic stages design in tester and now it is. Fixes: #12928 Differential Revision: [D79120574](https://our.internmc.facebook.com/intern/diff/D79120574/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D79120574 |
…ure, making it composable" RFC: #12660 This diff introduces composable architecture for the export pipeline. Changes: 1. Introduces notion of `PipelineArtifact` which encompasses stage artifacts and run context necessary to execute a stage and the pipeline. 1. Move all stages to new file to better architect the component 1. Export api will now accept `pipeline_stages` and honors the sequence if provided otherwise fallback to default sequence: `source_transform -> quantize -> torch.export -> to_edge_transform_and_lower -> to_executorch` 1. Validate if pipeline provided is valid, error out if not. 1. Add unittests to test pipeline sequence and stages With this one can execute a partial pipeline, for example, one can just do `TORCH_EXPORT -> TO_EDGE_TRANSFORM_AND_LOWER -> TO_EXECUTORCH` Current limitation: 1. `TORCH_EXPORT` stage is mandatory to avoid someone passing incorrect input such as `ExportedProgram` instead of nn.module, this enforcement will anchor the pipeline to have expected stages to run. - This limitation will be removed if we choose to add the support to take `ExportedProgram` as input. Note: 1. this diff is large because I moved stages to separate file and added bunch of unittests but core functionality that is added can be reviewed in `export.py` file. CC: digantdesai 2. Export component was supposed to mimic stages design in tester and now it is. Fixes: #12928 Differential Revision: [D79120574](https://our.internmc.facebook.com/intern/diff/D79120574/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D79120574 |
facebook-github-bot
merged commit d71c083
into
gh/abhinaykukkadapu/1/base
102 of 104 checks passed
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
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.
Stack from ghstack (oldest at bottom):
RFC: #12660
This diff introduces composable architecture for the export pipeline.
Changes:
PipelineArtifactwhich encompasses stage artifacts and run context necessary to execute a stage and the pipeline.pipeline_stagesand honors the sequence if provided otherwise fallback to default sequence:source_transform -> quantize -> torch.export -> to_edge_transform_and_lower -> to_executorchWith this one can execute a partial pipeline, for example, one can just do
TORCH_EXPORT -> TO_EDGE_TRANSFORM_AND_LOWER -> TO_EXECUTORCHCurrent limitation:
TORCH_EXPORTstage is mandatory to avoid someone passing incorrect input such asExportedPrograminstead of nn.module, this enforcement will anchor the pipeline to have expected stages to run.ExportedProgramas input.Note:
export.pyfile. CC: @digantdesaiFixes: #12928
Differential Revision: D79120574