8000 [docs] [onnx] Update the max admissible `opset_version` in the docs of `torch.onnx.export` and maybe update the default `opset_version` to 17 in release 2.1? · Issue #107446 · pytorch/pytorch · GitHub
[go: up one dir, main page]
Skip to content

[docs] [onnx] Update the max admissible opset_version in the docs of torch.onnx.export and maybe update the default opset_version to 17 in release 2.1? #107446

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
8000 Closed
vadimkantorov opened this issue Aug 18, 2023 · 2 comments
Closed

[docs] [onnx] Update the max admissible opset_version in the docs of torch.onnx.export and maybe update the default opset_version to 17 in release 2.1? #107446

vadimkantorov opened this issue Aug 18, 2023 · 2 comments
Labels
module: onnx Related to torch.onnx topic: docs topic category triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Comments

@vadimkantorov
Copy link
Contributor
vadimkantorov commented Aug 18, 2023 •
edited
Loading

🐛 Describe the bug

Nightly docs: https://pytorch.org/docs/master/onnx.html?highlight=export#torch.onnx.export
image

says that the max supported opset_version is 16, but searching through the issues one can see that #105134 even opset_version = 18 does not error out and often opset_version = 17 works

Is https://github.com/onnx/onnx/blob/main/docs/Operators.md#LayerNormalization supported by torch.onnx.export?

Updating the default opset_version to 17 can help to graph compilers, as more complex ops (like LayerNorm) are natively supported starting from opset_version = 17
@vadimkantorov vadimkantorov changed the title [docs] [onnx] Update the max admissible opset_version in the docs of torch.onnx.export [docs] [onnx] Update the max admissible opset_version in the docs of torch.onnx.export and maybe update the default opset_version to 17 in release 2.1? Aug 18, 2023
@ezyang ezyang added module: onnx Related to torch.onnx triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module topic: docs topic category labels Aug 18, 2023
@github-project-automation github-project-automation bot moved this to Inbox in ONNX Aug 18, 2023
@vadimkantorov vadimkantorov mentioned this issue Aug 21, 2023
Open
@vadimkantorov
Copy link
Contributor Author
8000
vadimkantorov commented Aug 21, 2023 •
edited
Loading

Also, because of this outdated default/confusion and issues in TensorRT, basic things like LayerNorm might not get subgraph-matched: NVIDIA/TensorRT#3230

This was referenced Aug 23, 2023
Closed
Closed
@justinchuby justinchuby mentioned this issue Aug 24, 2023
Closed
@github-project-automation github-project-automation bot moved this from Inbox to Done in ONNX Aug 24, 2023
@vadimkantorov
Copy link
Contributor Author
vadimkantorov commented Aug 24, 2023 •
edited
Loading

If torch.onnx.export is becoming legacy, it would be nice to have a warning-pane on top redirecting the users to the new dynamo_export? Or maybe best would be to unify legacy .export and .dynamo_export in a single function (e.g. .export(dynamo = True) with dynamo = False by default). And maybe make dynamo=True default or even make dynamo=False unsupported in a new major version of PyTorch (where e.g. older JIT/tracing can be removed from the codebase)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
module: onnx Related to torch.onnx topic: docs topic category triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
ONNX
Status: Done
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

[ONNX] Cap opset version at 17 for torch.onnx.export justinchuby/pytorch
2 participants
@ezyang @vadimkantorov
0