This option was originally added to work around pytorch/pytorch#141881. Now that that issue is fixed in PyTorch Core, we should deprecate the workaround in torchao. This is nice because this workaround has been blocking code simplification in Float8Linear.