nvidia

/

stt_pt_fastconformer_hybrid_large_pc

like 2

Follow

NVIDIA 58.6k

The model does not load when trying to finetune with the speech_to_text_finetune.py script from Nemo framework, due to empty values and incorrect handling of missing values in the code before they are overriden.
workaround:

1. untar .nemo file
2. edit the model_config.yaml
   replace '???' values in [train_ds: manifest_filepath, validation_ds:manifest_filepath, test_ds:manifest_filepath ] with

- - /some/path
replace '???' for tokenizer: dir path with
/some/path

3. tar files back to a .nemo file and use it