zhuchen03/FreeLB

Issues

• NaN encounted if FreeLB is used at the beginning of finetune stage

  #19 opened 3 years ago by BoomSky0416
  0

• Some confusion about the detach operation and embeds_init

  #18 opened 4 years ago by yupeijei1997
  0

• Errors generated during data preprocessing

  #17 opened 4 years ago by yupeijei1997
  2

• 词向量空间的不变性

  #16 opened 4 years ago by yupeijei1997
  0

• API Key

  #15 opened 4 years ago by nickboucher
  1

• A few questions about FreeLB and dropout

  #14 opened 4 years ago by zjiehang
  2

• the "dp_mask" problem

  #13 opened 4 years ago by Thornhill-GYL
  4

• Having issues with training RoBERTa. Loss not decreasing

  #12 opened 4 years ago by GodXuxilie
  2

• FreeLB-RoBERTa within HuggingFace's transformers?

  #2 opened 4 years ago by zhengwsh
  1

• ImportError: cannot import name 'glue_criterion_metrics' from 'transformers'

  #7 opened 4 years ago by SenYan1999
  6

• Reproducing results from the paper with roberta using fairseq

  #11 opened 4 years ago by bminixhofer
  5

• Is it still working with update_freq > 1?

  #4 opened 5 years ago by hitvoice
  2

• FreeLB didn't use the original training samples?

  #10 opened 5 years ago by YawYoung
  2

• Would you please release the hyper-parameters for FreeLB based on ALBERT(hugging-face)

  #9 opened 5 years ago by FFYYang
  6

• Does anyone meet the Nan error during the end epochs of training?

  #8 opened 5 years ago by PantherYan
  5

• Regarding the release of FreeLB ^_^

  #6 opened 5 years ago by None403
  0

• 'AlbertForSequenceClassification' object has no attribute 'encoder'

  #3 opened 5 years ago by trueto
  2

• Could you add some comments in the code？

  #1 opened 5 years ago by PengboLiu
  1