Megatron-DeepSpeed
Adding language specific validation sets for Multilingual model training
#97
Merged

Commits
  • adding multiple datasets to args
    hadyelsahar committed 4 years ago
  • Extra_valid_dataset arg & display on Tensorboard
    hadyelsahar committed 4 years ago
  • code cleaning
    hadyelsahar committed 4 years ago
  • code cleaning
    hadyelsahar committed 4 years ago
  • add run script example for multilingual validation
    hadyelsahar committed 4 years ago
  • rename extra-validation to periodic-eval
    hadyelsahar committed 4 years ago
  • bugfix + logging amount of training data per language
    TevenLeScao committed 4 years ago
  • Merge branch 'main' into main
    TevenLeScao committed 4 years ago
  • change "periodic-eval" to "extra-eval"
    hadyelsahar committed 4 years ago
  • elaborate tree sturcture in Multiling run script
    hadyelsahar committed 4 years ago
  • initialize all_x_datasets to None if empty
    hadyelsahar committed 4 years ago
  • clean kwargs.get
    hadyelsahar committed 4 years ago
  • add mode two of data loading
    hadyelsahar committed 4 years ago
  • adding option2 for data loading
    hadyelsahar committed 4 years ago
  • fix missing argument and range of split
    hadyelsahar committed 4 years ago
  • add multilingual run script
    hadyelsahar committed 4 years ago
  • small cleanups + fixed bugs with test iterator
    TevenLeScao committed 4 years ago
  • Fixed prefixlm
    TevenLeScao committed 4 years ago
  • adding variable cutoff for last epoch rather than 80%
    TevenLeScao committed 4 years ago
Loading