Add test for torch models tracing #2658
Conversation
lostella
added
tests
test/torch/model/test_jit.py
Outdated
| torch.zeros(shape, dtype=model.input_types()[name]) | ||
| for (name, shape) in model.input_shapes().items() |
There was a problem hiding this comment.
I think it would be nice to not have two different methods to get information on values.
There was a problem hiding this comment.
OK, I guess that can be addressed separately?
test/torch/model/test_jit.py
Outdated
| from gluonts.torch.model.tft import TemporalFusionTransformerModel | ||
|
|
||
|
|
||
| def get_model_and_input(model): |
There was a problem hiding this comment.
What's the purpose of this, why not just use .example_input_array?
Also, why should this be part of the parameter and not just be invoked in the test function?
There was a problem hiding this comment.
I'm not sure that using these completely dummy arrays is the best approach. Maybe for tracing the model yes, but then we should probably test the model vs scripted model using some other input
There was a problem hiding this comment.
So it's mostly a placeholder for now
There was a problem hiding this comment.
What's the purpose of this, why not just use .example_input_array?
That's not a property of torch.nn.Module
There was a problem hiding this comment.
Right. Still looks like something which could be consolidated.
lostella
changed the title
Description of changes: Add test to check that tracing a model results in the same computation, i.e., that the model's behaviour does not depend on input values.
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.
Please tag this pr with at least one of these labels to make our release process faster: BREAKING, new feature, bug fix, other change, dev setup