Avoid downloading sample input and output for inference

[qin-yu]

If I want to use core for inference, the model will be downloaded along with the sample input and output, which are unnecessary.

[FynnBe]

if you want to circumvent these additional downloads you can use perform_io_checks=False:

with ValidationContext(perform_io_checks=False):
    model_descr = load_model_description("affable-shark")

This will not download anything other than the rdf.yaml.
You can still use the resulting model description to run inference with core (it will then download weights on demand).

[qin-yu]

Thanks! Trying it.

[qin-yu]

Hi @FynnBe, I don't think with ValidationContext(perform_io_checks=False): is the right way to do it:

I used your fix in my PlantSeg branch, but it seems that everything is still downloaded when GitHub Action runs the tests, e.g. the "Run Tests with PyTest" part.

[qin-yu]

Could you explain what is the difference between:

with ValidationContext(perform_io_checks=False):  # method 1
    model_descr = load_model_description("affable-shark")

and

model = load_model_description(model_id, perform_io_checks=False)  # method 2

It seems that method 2 may work while method 1 doesn't.

[qin-yu]

I used your fix in my PlantSeg branch, but it seems that everything is still downloaded when GitHub Action runs the tests, e.g. the "Run Tests with PyTest" part.

I confirm that method 1 and 2 have different results, according to the new CI run.

[FynnBe]

oh sorry, yes using the context currently does not work for load_description (and load_model_description) as the default is set to the environment variable and not taken from the active context... sorry about that. I'll fix it in the next release!