[pytorch] Make behavior of SobolEngine consistent w/ other RNG functions (#36427)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36427

Addresses https://github.com/pytorch/pytorch/issues/36341

Test Plan: unit tests

Reviewed By: ldworkin

Differential Revision: D20952703

fbshipit-source-id: 28055f4c4c0f8012c2d96e473b822fa455dd833c

test/test_torch.py

@@ -3618,6 +3618,18 @@ class _TestTorchMixin(object):
         engine_3d.fast_forward(2)
         self.assertEqual(engine_3d.draw(5), actual_3d[5:])
 
+    def test_sobolengine_scrambled_lowdim_default_rng(self):
+        expected_1d = [0.039826, 0.484409, 0.953192, 0.799275, 0.267996]
+        torch.manual_seed(123456)
+        engine_1d = torch.quasirandom.SobolEngine(1, scramble=True)
+        actual_1d = engine_1d.draw(5)
+        self.assertEqual(actual_1d[:, 0], expected_1d)
+        torch.manual_seed(123456)
+        expected_3d = [0.133490, 0.480183, 0.855304, 0.970967, 0.345844]
+        engine_3d = torch.quasirandom.SobolEngine(3, scramble=True)
+        actual_3d = engine_3d.draw(5)
+        self.assertEqual(actual_3d[:, 0], expected_3d)
+
     def test_sobolengine_scrambled_highdim(self):
         engine = torch.quasirandom.SobolEngine(1111, scramble=True)
         draws = engine.draw(1000)

torch/quasirandom.py

@@ -57,11 +57,11 @@ class SobolEngine(object):
         torch._sobol_engine_initialize_state_(self.sobolstate, self.dimension)
 
         if self.scramble:
-            g = torch.Generator()
             if self.seed is not None:
+                g = torch.Generator()
                 g.manual_seed(self.seed)
             else:
-                g.seed()
+                g = None
 
             shift_ints = torch.randint(2, (self.dimension, self.MAXBIT), device=cpu, generator=g)
             self.shift = torch.mv(shift_ints, torch.pow(2, torch.arange(0, self.MAXBIT, device=cpu)))