Skip to content

Conversation

@pk-g
Copy link
Contributor

@pk-g pk-g commented May 1, 2019

@pytorchbot pytorchbot added oncall: jit Add this issue/PR to JIT oncall triage queue module: onnx Related to torch.onnx module: pybind Related to our Python bindings / interactions with other Python libraries labels May 1, 2019
@pk-g pk-g force-pushed the peykash/var_length_export branch 2 times, most recently from cc22369 to 0ff3c48 Compare May 3, 2019 22:00
@ezyang ezyang requested a review from houseroad May 6, 2019 14:39
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you mean when we use dynamic_axes, we must set input_names and output_names as well?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not exactly. User does NOT have to to provide input_names/ output_names on export API, if they already know what names are being used in the graph, they shall simply provide such names as keys to dynamic_axes dictionary and NOT specify input_names / output_names . however, if they do not know the names OR intend to use different names on exported model, they also have the option. They can do so by providing those names in input_names / output_names and using them in dynamic_axes.

@pk-g
Copy link
Contributor Author

pk-g commented May 13, 2019

@houseroad gentle reminder in case there are more feedback on this. thanks!

Copy link
Member

@houseroad houseroad left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we should ask users to specify the input_names/output_names if they use dynamic_axes. Otherwise, they have to guess the name of the inputs/outputs, because they don't have the control of them.

@pk-g pk-g force-pushed the peykash/var_length_export branch from 0ff3c48 to 67eeec3 Compare May 31, 2019 02:02
@pk-g
Copy link
Contributor Author

pk-g commented May 31, 2019

I think we should ask users to specify the input_names/output_names if they use dynamic_axes. Otherwise, they have to guess the name of the inputs/outputs, because they don't have the control of them.

@houseroad that's a good point. we are already letting user know by generating a warning during _validate_dynamic_axes method. The warning is essentially generated if names are not matching or they are not provided as input/output names.

@pk-g pk-g force-pushed the peykash/var_length_export branch 3 times, most recently from 7299d67 to f814e2d Compare June 3, 2019 18:07
@pk-g
Copy link
Contributor Author

pk-g commented Jun 3, 2019

@houseroad gentle reminder ... thanks!

Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@houseroad has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Copy link
Member

@houseroad houseroad left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@pk-g pk-g force-pushed the peykash/var_length_export branch from f814e2d to 05ae9d4 Compare June 4, 2019 17:21
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@houseroad has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

@facebook-github-bot
Copy link
Contributor

@houseroad merged this pull request in 98e3aae.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Merged module: onnx Related to torch.onnx module: pybind Related to our Python bindings / interactions with other Python libraries oncall: jit Add this issue/PR to JIT oncall triage queue open source

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants