Running pianoroll_rnn_nade_generate off of trained data fails with Key decoder/multi_rnn_cell/cell_2/basic_lstm_cell/kernel not found in checkpoint

Question

I was following the tutorial listed here: https://github.com/tensorflow/magenta/tree/master/magenta/models/pianoroll_rnn_nade

I successfully created a NoteSequences, a SequenceExamples, Trained the model and Evaluated.

When I get to the Generate Pianoroll Tracks section the application exits stating:

tensorflow.python.framework.errors_impl.NotFoundError: Key decoder/multi_rnn_cell/cell_2/basic_lstm_cell/kernel not found in checkpoint

I traced that error (or believed it was the right conclusion) to a thread discussing rnn refactoring in TF (https://github.com/tensorflow/tensorflow/issues/8447). They mention if you rename your variables in the checkpoint it will work but after examining my shape I don't seem to have any of the legacy vars (output of reader.get_variable_to_shape_map()):

{
'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/kernel/Adam_1': [216, 512],
'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/kernel/Adam': [216, 512],
'decoder/dense/bias/Adam_1': [216],
'global_step': [],
'nade/w_enc/Adam': [88, 1, 128],
'nade/w_enc': [88, 1, 128],
'decoder/dense/kernel/Adam_1': [128, 216],
'nade/w_dec_t/Adam_1': [88, 128, 1],
'nade/w_enc/Adam_1': [88, 1, 128],
'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/bias': [512],
'decoder/dense/bias': [216],
'decoder/dense/bias/Adam': [216],
'beta2_power': [],
'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/bias/Adam_1': [512],
'nade/w_dec_t/Adam': [88, 128, 1],
'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/bias/Adam': [512],
'nade/w_dec_t': [88, 128, 1],
'beta1_power': [], 'decoder/multi_rnn_cell/cell_0/basic_lstm_cell/kernel': [216, 512],
'decoder/dense/kernel': [128, 216],
'decoder/dense/kernel/Adam': [128, 216]
}

Does anyone have any idea what I missed? Has anyone successfully run the pianoroll_rnn_nade_generate model?