Deprecation warning thrown from use of old add_() signature; fix

[tristanengst]

It seems as though the usage of add_() in lars.py is deprecated. I get

home/tristanengst/miniconda3/envs/py39ISICLE/lib/python3.9/site-packages/torchlars/lars.py:140: UserWarning: This overload of add_ is deprecated:
        add_(Number alpha, Tensor other)
Consider using one of the following signatures instead:
        add_(Tensor other, *, Number alpha) (Triggered internally at  /opt/conda/conda-bld/pytorch_1639180487213/work/torch/csrc/utils/python_arg_parser.cpp:1050.)
  p.grad.add_(weight_decay, p.data)

whenever I use the optimizer. I believe the signature of add_() was changed in PyTorch 1.5, so we're well past a time when the old signature could represent a somewhat current PyTorch version.

Unless I'm mistaken, this can be fixed by changing line 140 of lars.py from

p.grad.add_(weight_decay, p.data)

to

p.grad.add_(p.data, alpha=weight_decay)

Would it be possible to see this change adopted for versions of torchlars installed alongside newer PyTorch versions? Thanks!

[RylanSchaeffer]

Has there been any update on this?

[YoojLee]

@RylanSchaeffer Not yet :(

[LiuYing-1]

Hi guys, does this lib still work even with the deprecation? thanks in advance.