| import torch | |
| class LinearMod(torch.nn.Linear): | |
| def __init__(self, *args, **kwargs): | |
| super().__init__(*args, **kwargs) | |
| def forward(self, input): | |
| return torch._C._nn.linear(input, self.weight, self.bias) | |
| print(torch.jit.trace(LinearMod(20, 20), torch.rand([20, 20])).graph) |