gen_replay_buffer_capacity VS demo_batch_size in tutorials code

[piteren]

In both tutorials: 3_train_gail.ipynb and 4_train_airl.ipynb
trainer demo_batch_size is higher than gen_replay_buffer_capacity, which produces gen_samples with many repetitions
(imitation/algorithms/adversarial/common.py #560 - e.g.for AIRL there are 2048 samples sampled from the buffer of size 512).
Is that intentional or a bug?