allow passing in progress=False for model_zoo.load_url call

[ssnl]

it eats up more than 40% lines of pytorch CI log

[fmassa]

We should do this!

[pmeier]

I would like to take that up (if that is oossible) in order to get to know the CI process better. I might need some guidance on where to start. Specifically, where is the code that needs improvement.

[fmassa]

@pmeier we would need to pass the progress argument of load_state_dict_from_url into all the call sites of model_zoo.load_url

vision/torchvision/models/resnet.py

Line 196 in ccbb322

model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))

I'm not yet clear on what is the best way to address this, given that the **kwargs are meant to be dispatched to the model, and not something else.

One possibility is to just add an extra progress=True to every call site there. I think it should not be BC-breaking, because the arguments that are passed to the models are kwargs only

[pmeier]

1. That part is clear. What I meant is: how do I set progress=False for the CI? Or did I misunderstand the problem?
2. Did you choose resnet as an example for all models here or is only applicable to it?

[ssnl]

@fmassa do we use master torchvision in pytorch ci?

@pmeier if the answer to the above question is yes, then after you make the change to torchvision, submit a pr to pytorch/pytorch and fix https://github.com/pytorch/pytorch/blob/5f82d59c0ae7d017bdb19a28b23fa1c19a0d6a9e/test/test_utils.py#L496-L515

[ssnl]

I'm not yet clear on what is the best way to address this, given that the **kwargs are meant to be dispatched to the model, and not something else.

One possibility is to just add an extra progress=True to every call site there. I think it should not be BC-breaking, because the arguments that are passed to the models are kwargs only

yes, progress=True, **kwargs should work.

[fmassa]

@ssnl not yet, but we will once #826 get's merged

[ssnl]

fixed via #875