[BUG] [flytekit] Task with torch.Tensor input fails when executing with flytekit remote access

[pimdh]

Describe the bug

Consider following task definition:

@task
def flyte_test_task(x: torch.Tensor) -> None:
    print(x)

Then register. Then execute with:

PROJECT = ...
DOMAIN = ...
remote = FlyteRemote(config=Config.auto())
task = remote.fetch_task(name="flyte_test_task", project=PROJECT, domain=DOMAIN)

execution = remote.execute(
    task,
    project=PROJECT,
    domain=DOMAIN,
    inputs={"x": torch.tensor([1, 2, 3])},
)

This triggers error:

TypeError: No automatic conversion found from type <class 'torch.Tensor'> to FlyteFile.Supported (os.PathLike, str, Flytefile)

Expected behavior

The execution should be created successfully.

Additional context to reproduce

Using flytekit version 1.14.3

Screenshots

No response

Are you sure this issue hasn't been raised already?

• Yes

Have you read the Code of Conduct?

• Yes

[kumare3]

This is a bug, we will take a look

[kumare3]

Notebook works - ofcourse you need torch installed in the image

[kumare3]

Ok, so the problem is type guessing resolves in the first available type - which is flytefile. We cannot use the type of the input directly as that may not be directly transferrable? @wild-endeavor do you know, we could simply use the type of the input especially for exotic types and always resolve correctly right?