sequence_mask on batches

[vilmar-hillow]

I've decided to run your code, and notebooks run just fine, with one-shot iterators. However, when I try to run trainer.py (which utilizes batching), I get the following error:
ValueError: maxlen must be scalar for sequence_mask
If I try to remove maxlen argument, I get another error instead:
TypeError: 'Tensor' object cannot be interpreted as an integer

Clearly sequence_mask is not ready for this batched input. How did you tackle this issue? I'm on tensorflow 1.4.