Use SVD with 2-D input (with batch_size) on GPU

I want to use ‘torch.svd()’ on GPU, but the funcation only accept 2-D input. I have implemented the svd by a for loop, but it’s too slow. In, it’s stated that “torch.svd” can be used in the same way as numpy.linalg.svd. But it causes an error when I pass a 3-D input to the function. Could you please help me?