Riva ASR Citrinet with WPE tokenizer

AI & Data Science Deep Learning (Training & Inference) Riva
1

Hi Team,

I am trying to depoy my NeMo model in Riva. I was able to convert .nemo to .riva. Then when I go to build step i am stuck at creating rmir file. The riva build step looks for tokenizer.model

I trained my nemo model using WPE and i have only vocab.txt. In the RIVA file generated, there is no field for tokenizer.model.

I see a bug report Bugreport: Riva ASR Citrinet with WPE tokenizer for this

Was this implemented? I do not see the status. Any other option available for me to use WPE tokenizer trained model in RIVA?

2

Hi @subalalitha

I will check regarding this issue with my team,

can you please share the riva build logs attached as file in this thread

Thanks

3

Hi @subalalitha

I have an update from the team,

The bug/ticket is still a feature request, and it will be developed over some time but not in near future

Thanks

4

Thanks for the update