search

microsoft / MASS

MASS: Masked Sequence to Sequence Pre-training for Language Generation
https://arxiv.org/pdf/1905.02450.pdf
Other
1.11k stars 206 forks source link

per-training BlUE always 0.0000 #158

Open Nanamumuhan opened 4 years ago

Nanamumuhan commented 4 years ago

@StillKeepTry when i per-training the model en-fr unsupNMT ,meet the same problem . How can i solve this problem image

jiaohuix commented 2 years ago

what's your pytorch、fairseq、cuda、gpu version？ i can't run MASS.