ghostnet_pytorch中的hardsigmoid无法转成onnx的hardsigmoid

[DamonsJ]

测试了ghostnet_pytorch中的模型，用的预训练的pth，转成onnx
结果hardsigmoid拆成了3个，不是单独的hardsigmoid是为什么呢？
这么转的onnx：
def main():
args = parser.parse_args()

model = ghostnet(num_classes=args.num_classes, width=args.width, dropout=args.dropout)
model.load_state_dict(torch.load('./models/state_dict_73.98.pth'))
args.num_gpu = 0
if args.num_gpu > 1:
    model = torch.nn.DataParallel(model, device_ids=list(range(args.num_gpu))).cuda()
elif args.num_gpu < 1:
    model = model
else:
    model = model.cuda()
print('GhostNet created.')

model.eval()
x=torch.randn((1,3,224,224))
torch.onnx.export(model, x, 'ghostnet.onnx', opset_version=11)

[iamhankai]

It's normal.