-
Notifications
You must be signed in to change notification settings - Fork 191
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
Conformer training with warp_rnnt #119
Comments
My configuration looks like this:
|
Sorry for late response. I'm very busy these days. First drop means that drop the sos token. It looks like a code correction is needed. But I've been busy lately so I can't afford it. |
What should be fixed? I could look into it. |
Sorry for too late response. How about train with ctc loss? (not transducer) I think there's a bug now, but I'm busy and can't afford to fix it. |
I've looked into training with CTC, but there was something wrong too - it crashed too, but now I do not have the error. I'll try to reproduce. |
I was wondering if you could comment why you're dropping 1st target here when training conformer with rnnt?
I'm asking because I got an error when doing the same with https://github.com/sooftware/conformer
where
criterion
iswarp_rnnt.rnnt_loss
.The error is
which makes sense when you look at the shapes:
The text was updated successfully, but these errors were encountered: