Modify round to support tracing

[maulikmadhavi]

Upon calling model tracing:

from ofa.model_zoo import ofa_net

ofa_network = ofa_net("ofa_mbv3_d234_e346_k357_w1.0", pretrained=True)
traced_model = torch.jit.trace(ofa_network, torch.randn(1, 3, 224, 224))

I run into the following error:

once-for-all/ofa/imagenet_classification/elastic_nn/modules/dynamic_layers.py in forward(self, x)
    225         if self.inverted_bottleneck is not None:
    226             self.inverted_bottleneck.conv.active_out_channel = make_divisible(
--> 227                 round(in_channel * self.active_expand_ratio),
    228                 MyNetwork.CHANNEL_DIVISIBLE,
    229             )

TypeError: type Tensor doesn't define __round__ method

I replaced round with torch.round.

Thanks
Maulik