Hey there @atrout!
I think @antgonza provided a pretty solid explanation for why you are experiencing this issue, but I will restate it here.
The feature IDs in your reference taxonomy artifact don't match the feature IDs in your reference sequences artifact. If I had to guess, your reference sequences here are actually not from the UNITE database, but are instead your representative sequence (the features have hashed sequences as the IDs, which is a default parameter for dada2 and deblur based denoising in QIIME 2). The reference sequences need to come from the same source as the reference taxonomy (those two parts add up to make the entire database, effectively). After importing, trimming, and then training, you can then introduce your representative sequences (the features identified in your study). Make sense?
Let us know, thanks! 
Have you had a chance to review the Training Feature Classifiers Tutorial?