[bug]: ipex install breaks non xpu devices

[saforem2]

It looks like this line:

Megatron-DeepSpeed/megatron/model/__init__.py

Line 4 in 61350c5

import intel_extension_for_pytorch

from #431 breaks things on non Intel systems.

A simple (not yet tested) fix for a device-angostic approach could be something like:

import torch

try:
    import intel_extension_for_pytorch as ipex
except Exception:
    pass

DEVICE = (
    "cuda" if torch.cuda.is_available() else (
        "xpu" if torch.xpu.is_available() else (
            "mps" if torch.backends.mps.is_available() else (
                "cpu"
            )
        )
    )
)

if DEVICE == 'cuda':
    from .fused_layer_norm import MixedFusedLayerNorm as LayerNorm
    from apex.normalization import MixedFusedRMSNorm as RMSNorm
else:
    if hasattr(torch.xpu, "IpexRmsNorm"):
        from .fused_rmsnorm import RMSNorm
    else:
        from .rmsnorm import RMSNorm
    from torch.nn import LayerNorm
# ...

which I believe should work

[vadam5]

I've also had this same issue recently. Can someone please fix this bug?

[clintg6]

+1