Float8Linear autocast's torch.get_autocast_gpu_dtype() is deprecated

[apaz-cli]

When I use:

from torchao.float8.float8_linear import Float8Linear, Float8LinearConfig
config = Float8LinearConfig()
self.w1 = Float8Linear(dim, hidden_dim, bias=False, config=config)

I get:

  /root/prime/.venv/lib/python3.10/site-packages/torchao/float8/float8_linear.py:430: DeprecationWarning: torch.get_autocast_gpu_dtype() is deprecated. Please use torch.get_autocast_dtype('cuda') instead. (Triggered internally at ../torch/csrc/autograd/init.cpp:787.)
    autocast_dtype = torch.get_autocast_gpu_dtype()

It also seems like the logic around autocasting around that point is incomplete in general. Lacking a CPU implementation.
https://github.com/pytorch/ao/blob/main/torchao/float8/float8_linear.py#L312

[jcaip]

Thanks for flagging! #1528 should fix the warning. I believe that float8 is only supported by GPUs currently. cc @vkuzo @drisspg if I'm mistaken