Correctly use test parameters (#166726)

This PR uses unused arguments in some tests.

Pull Request resolved: #166726
Approved by: https://github.com/rec, https://github.com/albanD, https://github.com/Skylion007

Author: cyyever

test/distributed/fsdp/test_fsdp_misc.py

@@ -514,18 +514,17 @@ def test_fsdp_optimizer_overlap(self):
     def test_fsdp_cpu_training(self):
         """Tests FSDP training on CPU."""
         gloo_pg = dist.new_group(backend="gloo")
-        for ss in [  # noqa: F841
+        for ss in [
             ShardingStrategy.NO_SHARD,
             ShardingStrategy.FULL_SHARD,
             ShardingStrategy.SHARD_GRAD_OP,
-            ShardingStrategy.HYBRID_SHARD,
-            ShardingStrategy._HYBRID_SHARD_ZERO2,
         ]:
             torch.manual_seed(42)
             model = MyModel()
             ref_model = DDP(deepcopy(model), process_group=gloo_pg)
             model = FSDP(
                 model,
+                sharding_strategy=ss,
                 auto_wrap_policy=always_wrap_policy,
                 process_group=gloo_pg,
                 device_id=torch.device("cpu"),

test/quantization/core/test_quantized_tensor.py

@@ -1062,8 +1062,8 @@ def _test_qtensor_masked_fill(self, device):
         mask = torch.randint(0, 2, (numel, ), device=device)
         mask = mask.bool()
         x = torch.rand(numel, device=device)
-        qx = torch.quantize_per_tensor(x, scale=scale, zero_point=zero_point, dtype=qtype)
         for qtype, fill_with in itertools.product(types, fills):
+            qx = torch.quantize_per_tensor(x, scale=scale, zero_point=zero_point, dtype=qtype)
             q_masked_fill = qx.clone()
             q_masked_fill.masked_fill_(mask, fill_with)
             ref = qx.clone()

test/test_spectral_ops.py

@@ -481,7 +481,7 @@ def test_fftn_noop_transform(self, device, dtype):
             torch.fft.ifft2,
         ]:
             inp = make_tensor((10, 10), device=device, dtype=dtype)
-            out = torch.fft.fftn(inp, dim=[])
+            out = op(inp, dim=[])
 
             expect_dtype = RESULT_TYPE.get(inp.dtype, inp.dtype)
             expect = inp.to(expect_dtype)

test/test_torch.py

@@ -2263,8 +2263,8 @@ def check(t, correction=1, fweights=None, aweights=None):
             if num_observations > 0:
                 fweights = torch.randint(1, 10, (num_observations,), device=device)
                 aweights = make_tensor((num_observations,), dtype=torch.float, device=device, low=1)
-                for correction, _fw, _aw in product([0, 1, 2], [None, fweights], [None, aweights]):
-                    check(x, correction, fweights, aweights)
+                for correction, fw, aw in product([0, 1, 2], [None, fweights], [None, aweights]):
+                    check(x, correction, fw, aw)
 
     @skipIfNoSciPy
     @dtypes(*floating_types_and(torch.half, torch.bfloat16))