each model is trained for the specified iterations

across all models:
 LR = 0.005
 CONTEXT = 6
 HIDDEN = 256

using NO punctuation tokenizer

random seed is 1234

final average losses:
 untrained = 6.987885
 10k  = 5.628099 
 50k  = 5.014620
 100k = 4.293952
 250k = 2.538245
 500k = 0.901952
 1mil = 0.268483
 2mil = 0.176046
 3mil = 0.149419