unsorted V.S. sorted example lengths file

#66
by allenxiao - opened

Thanks for providing this great tool.
I notice the example 'pretraining_new_model''s parameter 'example_lengths_file' has been replaced with the unsorted version. So what impact would such a change have for the model? Does it mean that the sorted version is a wrong parameter? If not, what is the difference for the model output?

Thank you for your question. Please check the discussion in closed issue #61. The lengths file gives the trainer information for padding.

ctheodoris changed discussion status to closed

Sign up or log in to comment