TensorFlowASR icon indicating copy to clipboard operation
TensorFlowASR copied to clipboard

[Question] Training for conformer model

Open SEOLJINYOUNG opened this issue 1 year ago • 2 comments

I'm trying to learn a Conformer model using data and a custom word dictionary. My learning environment uses two GPUs, and the dataset uses commonvoice data.

Looking at this log, can i see that learning is progressing normally?

image

image

And do I need to see the log below? INFO:tensorflow:Collective all_reduce tensors: 1 all_reduces, num_devices = 2, group_size = 2, implementation = CommunicationImplementation.AUTO, num_packs = 1

Thank you in advance for your response.

SEOLJINYOUNG avatar Nov 11 '24 02:11 SEOLJINYOUNG

您好,您的邮件我已收到。我会尽快给您回复。祝好!

Aegon007 avatar Nov 11 '24 02:11 Aegon007

@SEOLJINYOUNG I see you use conformer-ctc, the conformer-ctc are currently suffering the nan loss problem and I'm still investigating that Meanwhile, you can try conformer-transducer model

nglehuy avatar Nov 13 '24 14:11 nglehuy