Fine-Tuning Conformer CTC Model

[AdityaYadavalli1]

Hi, I am new to NeMo and I've been trying to fine-tune the stt_en_conformer_ctc_large model on the common voice Japanese dataset by following this tutorial. Instead of loading stt_en_citrinet_512 as shown in the Subword Encoding CTC Model section, I changed it to stt_en_conformer_ctc_large. After this change, when I run model.setup_training_data(cfg.train_ds), I get the following error:

manifest_filepaths (length=1) and tarred_audio_filepaths (length=8) need to have the same number of buckets.

How do I fix this?

[titu1994]

I think the tutorial prints model.cfg.train_ds, so just lookup if it has us_tarred and set that to false

[AdityaYadavalli1]

Thanks! It was a silly typo on my end but you pointed in the direction.