TensorFlowASR icon indicating copy to clipboard operation
TensorFlowASR copied to clipboard

'SubwordFeaturizer' object has no attribute 'vocab_array'

Open kkiyani opened this issue 4 years ago • 8 comments

testing Jasper using subwords

new_error

kkiyani avatar May 25 '21 09:05 kkiyani

@kkiyani I haven't tried Subword for CTC models but the PR should work

nglehuy avatar May 26 '21 16:05 nglehuy

PR?

kkiyani avatar May 26 '21 17:05 kkiyani

@kkiyani It's the pull request, you can pull the newest commit on main to test 😄

nglehuy avatar May 26 '21 17:05 nglehuy

Thank you so much.

kkiyani avatar May 26 '21 17:05 kkiyani

One more question: The snapshot for training session doesn't get save in the case of contextnet

kkiyani avatar May 26 '21 17:05 kkiyani

One more question: The snapshot for training session doesn't get save in the case of contextnet

@kkiyani Could you be more specific?

nglehuy avatar May 26 '21 18:05 nglehuy

It says it's not functional and sequential model, it's a subclass model. If you want to save the model use save_weight() Or save_format='tf'

kkiyani avatar May 26 '21 18:05 kkiyani

It says it's not functional and sequential model, it's a subclass model. If you want to save the model use save_weight() Or save_format='tf'

If you want to save the whole model in h5 format, then you'll have to define the get_config method of every layer and model classes, otherwise you'll have to go with the tf format.

nglehuy avatar May 26 '21 18:05 nglehuy

I’ll close the issue here. Feel free to reopen if you have further questions.

nglehuy avatar Sep 02 '22 05:09 nglehuy