when training in Race , The eval_accuracy is flat , it only has three numbers which are 0.0, 0.33334, 0.66667, 1.0 - Githubissues

google-research / albert

ALBERT: A Lite BERT for Self-supervised Learning of Language Representations

Apache License 2.0

3.23k stars 571 forks source link

when training in Race , The eval_accuracy is flat , it only has three numbers which are 0.0, 0.33334, 0.66667, 1.0 #264

Open AGuangzzz opened 2 years ago