Tensodrboard values

[AlexandderGorodetski]

Hello guys,

I have a question regarding tedlium3 recipe tensorboard.

What are meanings of the following values:

current_loss current_pruned_loss current_simple_loss tot_loss

Thanks very much, AlexG.

[csukuangfj]

https://github.com/k2-fsa/icefall/blob/6693d907d3ddd5c5eade144b55a57c8831d6d9b2/egs/librispeech/ASR/pruned_transducer_stateless7/train.py#L695-L698

current_loss is the loss of the current batch, which is computed by https://github.com/k2-fsa/icefall/blob/6693d907d3ddd5c5eade144b55a57c8831d6d9b2/egs/librispeech/ASR/pruned_transducer_stateless7/train.py#L685

current_pruned_loss is the pruned loss of the current batch. current_simple_loss is the simple loss of the current batch. The above two losses are computed by https://github.com/k2-fsa/icefall/blob/6693d907d3ddd5c5eade144b55a57c8831d6d9b2/egs/librispeech/ASR/pruned_transducer_stateless7/train.py#L662

tot_loss is the total loss averaged over all batches that have seen so far (with some smoothing).