torchao.float8 not working on PyTorch 2.4.1 and how does torchao handle FP8 autocast?

We know that `Transformer_Engine` provides context manager named [fp8_autocast](https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/examples/fp8_primer.html#FP8-autocasting) to handle autocasting.

But I cannot find a similar context manager in `torchao`, so could anybody provides some hint about how does `torchao` handle FP8 autocasting? 

Thanks!