Evaluation function should be changed to use a Seq2Seq object.
Evaluation function should take a list of AccuracyMeter objects. We should support
Sentence level accuracy
Token level accuracy
first word accuracy
length accuracy
Each of these should have a class that subclasses from AccuracyMeter. It should save them to the cox store and optionally print a summary to STDOUT
Evaluation function should be changed to use a Seq2Seq object. Evaluation function should take a list of AccuracyMeter objects. We should support