consume long time for predicting validation output about nmt-keras HOT 3 CLOSED

It seems something is wrong. Can you give me more information? Are you using GPU? Can you share your config file?

from nmt-keras.

yes I use GPU on google colaboratory, I am using the default configuration: https://github.com/lvapeab/nmt-keras/blob/master/config.py but
MAX_INPUT_TEXT_LEN = 300,
Use_Cuda= True
these are different parameters
if I make validation data = 100 sample, this makes it fast.
but my valid is 7000 too long time

from nmt-keras.

Indeed, the validation time is linearly proportional w.r.t. the number of samples (currently, no batch decoding is implemented). So you may want to validate on a smaller set (but big enough to take generalizable takeaways). This time is also proportional to the number of max steps allowed in the beam search (MAX_OUTPUT_TEXT_LEN_TEST). If the model hasn't learned yet to generate sentences of the adequate length, it will take longer, as the search will make this number of steps.

You should also take a look to your data and check whether it makes sense setting MAX_INPUT_TEXT_LEN=300. If you are doing regular word-level MT, this value seems too high.

from nmt-keras.