allowing parameters to be initialised from a list

[kathryn-baker]

This resolves the following error when initialising parameters from a JSON file which by default uses lists rather than tensors:

File [~/notebooks/calibration-training/training/calibration_modules/decoupled_linear.py:154], in DecoupledLinearInput.__init__(self, model, x_size, x_mask, **kwargs) ...
--> [145]         value = float(value) * torch.ones(size)
    [146]     value_size = value.shape
    [147]     if value.dim() == 1 and isinstance(size, int):

TypeError: float() argument must be a string or a real number, not 'list'

[t-bz]

I think we should definitely support this. Could you rebase to fix the CI tests and add a test similar to this?

[kathryn-baker]

I've added the test you suggested and pulled in your changes from the main branch but I seem to get a failing test for something I didn't modify to do with test_save_and_load(), is it possible it's a platform issue?

[t-bz]

It seems like the test passed on the same platform before. I'll look into it if it keeps reoccurring.