This is more or less a straight translation of Andrej Karpathy's Recurrent Neural Network code from Python to Go. See http://karpathy.github.io/2015/05/21/rnn-effectiveness for more information.
I have attempted to translate it faithfully, even down to the level of preserving variable names (many of which are somewhat... terse) and his comment text. The one major change I did introduce is code for checkpointing the model; this is primarily implemented in persistence.go.
Any errors here are my own, and not Karpathy's. Corrections welcome.
$ go get github.com/armhold/rnn/...$ cd $GOPATH/src/github.com/armhold/rnn$ rnn -i input.txt
This will run the network on a small corpus of Shakespeare text. After a few thousand iterations, you should start seeing output that looks superficially like a Shakespeare play.