Repository navigation
Add Native runtime to torchvision export workflow (#24286) - #24286
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24286
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 39a5b5e with merge base 043c65b ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@JakeStevens has exported this pull request. If you are a Meta employee, you can view the originating Diff in D128174534. |
This PR needs a
|
6569bf6 to
35cae88
Compare
Summary: Pull Request resolved: pytorch#24286 Add `--runtime native` to `export_torchvision`. It lowers the capture with `to_native()` and standard Native preprocessing, writes `model.ptn`, and records Native provenance in `export.json`. The saved input and eager logits match the ET path, so `--pt2` can reuse one capture to compare both pipelines. Authored with OpenAI Codex. Differential Revision: D128174534
Summary: Add `--runtime native` to `export_torchvision`. It lowers the capture with `to_native()` and standard Native preprocessing, writes `model.ptn`, and records Native provenance in `export.json`. The saved input and eager logits match the ET path, so `--pt2` can reuse one capture to compare both pipelines. Authored with OpenAI Codex. Differential Revision: D128174534
35cae88 to
39a5b5e
Compare
digantdesai
left a comment
There was a problem hiding this comment.
Review automatically exported from Phabricator review in Meta.
Summary:
Add
--runtime nativetoexport_torchvision. It lowers the capture withto_native()and standard Native preprocessing, writesmodel.ptn, and records Native provenance inexport.json. The saved input and eager logits match the ET path, so--pt2can reuse one capture to compare both pipelines.Authored with OpenAI Codex.
Differential Revision: D128174534