GPU Peak performance calculation requires FP8 and GH200 support

[le1nux]

Currently, the GPU peak performance calculation is rather limited supporting only BF16 and FP16 precision, as well as, A100 and H100 GPUs only.

modalities/src/modalities/utils/mfu.py

Line 96 in ef2f8c2

def _get_theoretical_gpu_peak_performance(wrapped_model: FSDPX, world_size: int) -> Optional[Number]:

We should extend this to support FP8 and GH200.