Training loss becomes constant at around 12.xxxx - Githubissues

baidu-research / ba-dls-deepspeech

Apache License 2.0

486 stars 174 forks source link

Training loss becomes constant at around 12.xxxx #29

Open faruk-ahmad opened 7 years ago

faruk-ahmad commented 7 years ago

While training the model initially training loss decreases but it remains constant at around 12 point some values. From 11000 to 20000 epoch, still it is in the same state with same loss. What might be the cause?