-
Notifications
You must be signed in to change notification settings - Fork 1
Language model example
The first example you might like to tinker with is a character-level language model. It is largely based on Andrej Karpathy blog post The Unreasonable Effectiveness of Recurrent Neural Networks.
First of all you will need a dataset. This can be any text file. You can just grab the Shakespeare dataset from Andrej Karpathy: https://github.com/karpathy/char-rnn/blob/master/data/tinyshakespeare/input.txt.
Now you need to convert your dataset into training batches. Run LanguageModel.exe Prepare -i=[path to your dataset]
and wait for batches to finish generating. You will find a new file with the same name as your training set and .bin
extension. Copy the path to this file.
Finally run LanguageModel Learn -b=[batches path]
substituting the path to the bin file. Hooray! You are learning stuff!
You can also specify -gpu
flag to use GPU and -gui
flag to show some nice GUI.
TODO