Allow data intake both as a single large .txt file or as a folder of .txt files.

Auto-convert all Unicode intake to ASCII, with additional flag to disable. Unicode (specifically, the processes under unidecode) may be responsible for tensor-length mismatch bug. Investigate.

See def random_training_set(chunk_len, batch_size).

UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number

For each saved model, find a way to store the training arguments - either in the title, or in an associated specs.txt file.

Create a simple web app that uses a saved model trained on a publicly available data such as Shakespeare.

$HOME/torch/install/bin/th train.lua -input_h5 $WORK/rap/rap.h5 -input_json $WORK/rap/rap.json -rnn_size 2048 -num_layers 3 -dropout 0.25 -max_epochs 65536 -seq_length 256 -batch_size 128 -checkpoint_name $WORK/rap_checkpoints/rap > output.txt

Also make CUDA default.

Current data set is composed of 1093 scripts in Unicode with strange tabbing / spacing. Would be nice to find a more processed data set.