Open
Description
📚 The doc issue
I think it would useful to expand AOTI tutorial to cover AMP trained models.
https://pytorch.org/tutorials/recipes/torch_export_aoti_python.html
Suggest a potential alternative/fix
No response
cc @svekars @brycebortree @sekyondaMeta @ezyang @chauhang @penguinwu @avikchaudhuri @gmagogsfm @zhxchen17 @tugsbayasgalan @angelayi @suo @ydwu4 @desertfire @chenyang78
Metadata
Metadata
Assignees
Labels
Type
Projects
Milestone
Relationships
Development
No branches or pull requests
Activity
bhack commentedon Nov 6, 2024
/cc @agunapal @angelayi @svekars
agunapal commentedon Nov 6, 2024
@bhack Thanks for the suggestion. Will check
svekars commentedon Nov 8, 2024
Let's move this to the tutorials repo.
bhack commentedon Jan 31, 2025
What is the status of this with 2.6.0 ?
angelayi commentedon Apr 16, 2025
@bhack what sort of model are you looking at? Could you provide a pointer?
I would think that an AMP trained model behaves the same as just any model that you pass to export?
bhack commentedon Apr 16, 2025
Nothing specific just a general guide how to export/aoti compile an amp model.
Mainly not only specifically amp trained but also the case when we want to use amp at inference.
E.g. do we need to amp wrap the model before export/compile? Etc..