Method f

test/test_proxy_tensor.py:151–153

(a)

    def test_pre_dispatch_mode_stack(self):
        def f(a):
            b = torch.ones(4, 4)
            return torch.matmul(a, b)
        # We expect to see matmul in the trace - it should NOT be decomposed into mm.
        # Also, torch.ones() doesn't show up in the trace.
        # This is annoying but expected: ones() never dispatches to the Autograd dispatch key,