OpenNMT

ValueError: invalid literal for int() with base 10

i after i tokenize voc. then i replace("_","’’) to delete all _ in voc. But when i start to train this error happend. ValueError: invalid literal for int() with base 10 vệ\t821670.
i just replace bảo_vệ 821670 → bảo vệ 821670. How to fix this