IterableDataset.with_format("torch") not working

[jiangwangyi]

Describe the bug

After calling the with_format("torch") method on an IterableDataset instance, the data format is unchanged.

Steps to reproduce the bug

from datasets import IterableDataset
def gen():
    for i in range(4):
        yield {"a": [i] * 4}
dataset = IterableDataset.from_generator(gen).with_format("torch")
next(iter(dataset))

Expected behavior

{"a": torch.tensor([0, 0, 0, 0])} is expected, but {"a": [0, 0, 0, 0]} is observed.

Environment info

platform==ubuntu 22.04.01
python==3.10.9
datasets==2.11.0

[lhoestq]

Hi ! Thanks for reporting, I'm working on it ;)