I have 30,000 parallel sentences. How many training steps I have to set? I am using transformer.
These two values are not necessarily strictly correlated. I encourage you to set a big value for train_steps
, and regularly save and evaluate checkpoints with your metrics (e.g. BLEU).
Side note: 30k parallel sentences is quite a small dataset. Depending on the task you might not get very good results.
1 Like
okay. Thanks a lot