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
[WIP] E-Branchformer Encoder in ESPnet2 #4812
Conversation
x_input (Union[Tuple, torch.Tensor]): Input tensor w/ or w/o pos emb. | ||
- w/ pos emb: Tuple of tensors [(#batch, time, size), (1, time, size)]. | ||
- w/o pos emb: Tensor (#batch, time, size). | ||
mask (torch.Tensor): Mask tensor for the input (#batch, time). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The shape of mask
is (#batch, 1, time)
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi @kkim-asapp Is the shape correct?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks, @pyf98! You're right. It was wrong. Let me update it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I remember it was wrong in Conformer (or other encoders). If so, can you update it as well?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sure, will do.
Looks good to me! I'm looking forward to seeing the new results and models. Since the |
@kkim-asapp, thanks a lot for the great PR! |
Right! Currently, I didn't touch any parts of |
I did! |
espnet/nets/pytorch_backend/conformer/contextual_block_encoder_layer.py |
You can run some of the test scripts locally instead of checking it via CI. |
I did, but the error logs are different. Let me check after reinstall dependencies. |
Oh, OK. |
Codecov Report
@@ Coverage Diff @@
## master #4812 +/- ##
===========================================
+ Coverage 69.06% 80.44% +11.38%
===========================================
Files 526 534 +8
Lines 46125 47118 +993
===========================================
+ Hits 31858 37906 +6048
+ Misses 14267 9212 -5055
Flags with carried forward coverage won't be shown. Click here to find out more.
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Hi @pyf98, is there anything I can do to fix one last failing check? Maybe I need to fix |
You don't need it. |
Please make a follow-up PR to provide an updated README file etc. |
Hi,
This PR is about the E-Branchformer encoder (Kim et al., SLT 2022: will be appeared, Arxiv: link).
The module is implemented in ESPnet2 with the base recipe for Librispeech ASR.
Also, I've adding an argument in
Repeat
module to support LayerDrop (link).This PR is in progress, and I will update results on README file.