search

thunlp / ELLE

32 stars 2 forks source link

Error: train.py: error: argument --task: invalid choice: 'continual_KI' #3

Closed Vincent950129 closed 2 years ago

Vincent950129 commented 2 years ago

Hi,

Thank you for the work. I am trying to run ELLE for pre-training with the script train_base_prompt.sh in fairseq_ELLE\examples\roberta. It complains

train.py: error: argument --task: invalid choice: 'continual_KI' (choose from 'sentence_prediction', 'language_modeling', 'speech_to_text', 'translation', 'multilingual_translation', 'semisupervised_translation', 'audio_pretraining', 'multilingual_masked_lm', 'cross_lingual_lm', 'denoising', 'multilingual_denoising', 'translation_from_pretrained_xlm', 'legacy_masked_lm', 'translation_multi_simple_epoch', 'masked_lm', 'translation_from_pretrained_bart', 'sentence_ranking', 'translation_lev', 'dummy_lm', 'dummy_masked_lm', 'dummy_mt')

Which one I should choose in order to continual pre-train?

Thank you for your time.

Neo-Zhangjiajie commented 2 years ago

Could you try adding this commond before the bash: export PYTHONPATH=fairseq_ELLE You need to guarantee that you fairseq libary is from fairseq_ELLE. Similarly, when you run code in fairseq-0.9.0, you need this commond: export PYTHONPATH=fairseq-0.9.0