Build an nn.BatchNorm1d whose running statistics and affine parameters are set to the tensors given, put it in evaluation mode, and apply it to x of shape (B, D).

Return the output. In evaluation mode BatchNorm normalises with its stored running statistics rather than the batch's own, and must not update them -- which is why the mode matters more here than almost anywhere else.

Input

x =
tensor([[1., 2.],
        [3., 4.]])
running_mean = tensor([1., 1.])
running_var = tensor([4., 1.])
weight = tensor([2., 1.])
bias = tensor([0., 1.])
eps = 1e-05

Output

tensor([[0.0000, 2.0000],
        [2.0000, 4.0000]])