-
Notifications
You must be signed in to change notification settings - Fork 74k
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.
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
[TF:TRT] Enable dynamic batch dim for Conv2dBackpropInput #51468
[TF:TRT] Enable dynamic batch dim for Conv2dBackpropInput #51468
Conversation
Thanks @christopherbate for this PR! I had a quick look at the changes, it looks good overall. |
This is what you have for the PR description: Full dynamic shape for height and width will come later. We first need to add an additional dynamic padding operation support in order to handle the case where the input_shapes operation argument is non-constant. Here is my suggestion to revise it. Do you see any problem? We enable dynamic batch dimension for Conv2DBackpropInput simply by relaxing the check to allow dynamic batch dimension in the converter. We also add positive and negative tests. The support for dynamic height and width will come later. This is because we will need to add an additional dynamic padding operation in order to support dynamic height and width for Conv2dBackpropinput. |
No problem, I updated the description. |
f92cfac
to
2355bd3
Compare
318ddf7
to
526ef46
Compare
526ef46
to
638e3f3
Compare
I updated test parameter naming to be inline with other dynamic shape developments. I've rebased on master, but it still it shows difference between |
This PR enables dynamic batch dimension for Conv2dBackpropInput.
We enable dynamic batch dimension for Conv2DBackpropInput simply by relaxing the check to allow dynamic batch dimension in the converter. We also add positive and negative tests.
The support for dynamic height and width will come later. This is because we will need to add an additional dynamic padding operation in order to support dynamic height and width for Conv2dBackpropinput.