daniel

What would be the best way to show/extract the confidence of the recognition (greedy or beam search) during inference. The sequence/word confidence would be interesting to see.

I ran into an issue following the readme steps: [0:00:07.825803][Epoch: 0][Iter: 58][Loss: 8.262327][lr: 0.000500]: 100%|██████████████████████████████████████████████████████████████████████████| 58/58 [00:08