Regarding hidden state size of GRU

[andyjung12]

In the paper, it said that the hidden state size of GRU for inter-epoch level of raw data is 64. But in the config that you used in the github here says that it is 256. Which one is correct?

[pquochuy]

There are two of of them, 1 for seqsleepnet branch and 1 for cnn-rnn brach. The former is one is 64, the latter is 256.

[andyjung12]

so, in the paper it said, "The LSTM and GRU cells responsible for inter-epoch sequential modelling had 256 and 64 units in their hidden state vectors, respectively"

It is other way around, right? 64 for seqsleepnet and 256 for cnn-rnn branch?

[pquochuy]

Yes, thats right.