You can estimate 1 byte == 1 char, so there are 17150000 chars in there xD
and I'm gonna take chat data into the dataset in order to make it improve while it's being used
Anyways, front end!
I'm gonna start the Word2vec script first and hope my computer doesn't run out of memory (if that happens, I'll be forced to shut down and it takes ages to get it back)
It converts text to tokens by comparing it to the ID's stored in a dictionary/vocabulary depending on naming (doesn't matter really :D), which what the W2V script does - it converts the training data to vectors/tokens/whatever
the vector form is inputted into the network (which is an RNN with LSTM meaning it is made for sequences), and the output is (in the same format) numbers in order
the result is then converted from the numbers back to text
and the main training and network is in the seq2seq script