[Data] Ray fails when using map_groups function and Gpus as Actors

[shivraj95]

Description

When using map_groups and setting the arg num_gpus > 0, Ray will raise an exception:

ValueError(
"batch_size must be provided to map_batches when requesting GPUs. "
"The optimal batch size depends on the model, data, and GPU used. "
"It is recommended to use the largest batch size that doesn't result "
"in your GPU device running out of memory. You can view the GPU memory "
"usage via the Ray dashboard."
)

from this code block.

Versions / Dependencies

This bug will occur for all Python and Ray versions as of post date.

Reproduction script

This must be run on a cluster with GPU workers.

ds = ray.data.from_items([ 
    {"group_id": 1, "value": 1},
    {"group_id": 1, "value": 2},
    {"group_id": 2, "value": 3},
    {"group_id": 2, "value": 4}])

ds.groupby('group_id').map_groups(lambda x: {"result": np.array([g["value"][0]])}, num_gpus=1)

Issue Severity

High: It blocks me from completing my task.

[bveeramani]

Ah, yeah. I was able to reproduce this.

Shouldn't be too hard to fix.

[shivraj95]

Great, thanks! Look forward to the fix :)

[anyscalesam]

@bveeramani ETA on fix? cc @c21