Essentially, my training log returns:
***** Running Evaluation *****
Num examples: Unknown
Batch size = 8
Saving model checkpoint to ./whisper-small/checkpoint-1000
Configuration saved in ./whisper-small/checkpoint-1000/config.json
Configuration saved in ./whisper-small/checkpoint-1000/generation_config.json
Model weights saved in ./whisper-small/checkpoint-1000/pytorch_model.bin
Feature extractor saved in ./whisper-small/checkpoint-1000/preprocessor_config.json
tokenizer config file saved in ./whisper-small/checkpoint-1000/tokenizer_config.json
Special tokens file saved in ./whisper-small/checkpoint-1000/special_tokens_map.json
added tokens file saved in ./whisper-small/checkpoint-1000/added_tokens.json
Step Training Loss Validation Loss Wer
1000 0.337200 0.387359 38.595348
2000 0.305800 0.368565 36.393789
3000 0.301600 0.333480 50.099939
4000 0.289900 0.332969 66.476927
5000 0.276800 0.327825 84.689030
6000 0.257700 0.316814 102.301498
Is it possible for me to print the expected text and the text generated at each eval step?
The WER looks super high but I doubt it’s because of special characters and I just want to make sure.