Why comparision of single value torch.Tensor with float - performance

1

Why comparison of single value tensor with float is much slower than a comparison of tensors item method output with float?
I am optimizing my code and have just seen this behavior. Is it expected?
Example below:

In [7]: one = torch.tensor(1.)

In [8]: %timeit one > 1.5
6.28 µs ± 9.48 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [9]: %timeit one.item() > 1.5
562 ns ± 1.18 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [19]: %timeit torch.gt(one, 1.5)
6.42 µs ± 18.9 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)
2

The tensor comparison would dispatch to internal PyTorch methods, might cast the dtype of the value you are comparing against and could even move it to the same device where the tensor is stored. The item() call creates a plain Python value to compare against the other float.