`trainer.predict()` uses incorrect batch size with an `IterableDataset`

[EricWiener]

🐛 Bug

test_iter = RandomIterableDataset(32, 10)
test_iter = DataLoader(test_iter, batch_size=32)

# Using an iterable dataset uses the incorrect batch size
trainer.predict(dataloaders=test_iter)

will result in the trainer using a batch size of 1 instead of 32. This issue only appears for IterableDataset (not Dataset).

To Reproduce

Notebook here

Expected behavior

The batch size specified in the DataLoader should be used.

Environment

* CUDA:
	- GPU:
		- Tesla T4
	- available:         True
	- version:           10.1
* Packages:
	- numpy:             1.18.5
	- pyTorch_debug:     False
	- pyTorch_version:   1.6.0+cu101
	- pytorch-lightning: 0.10.0
	- tqdm:              4.41.1
* System:
	- OS:                Linux
	- architecture:
		- 64bit
		- 
	- processor:         x86_64
	- python:            3.6.9
	- version:           #1 SMP Thu Jul 23 08:00:38 PDT 2020

Additional context

This used to work before 1.5.1 (not sure exactly which version broke it though).

cc @rohitgr7

[tchaton]

Hey @EricWiener,

Would you be interested in making a PR to resolve this bug?

Best,
T.c

[EricWiener]

Hey @tchaton, I would love to, but very low on time for the remainder of the month. Might take me a while to get to it, but if no one gets to it earlier I can do it.