[7/N] Fix unused loop variables in tests (pytorch#167043)

This PR continues to fix or remove unused loop variables in tests.

Pull Request resolved: pytorch#167043
Approved by: https://github.com/Lucaskabela

Author: cyyever

test/dynamo/test_compiler_bisector.py

@@ -283,7 +283,7 @@ def test_fn():
     )
     def test_bisect_pre_grad_graph(self):
         def f(x):
-            for i in range(5):
+            for _ in range(5):
                 x = x + 1
             return x.relu()
 

test/functorch/test_control_flow.py

@@ -942,9 +942,7 @@ def false_fn(x):
         b = torch.randn(4, requires_grad=True)
         c = torch.randn(4, requires_grad=True)
 
-        for pred, fn in zip(
-            [torch.tensor(False), torch.tensor(True)], [false_fn, true_fn]
-        ):
+        for pred in [torch.tensor(False), torch.tensor(True)]:
             with self.assertRaisesRegex(
                 torch._dynamo.exc.UncapturedHigherOrderOpError,
                 "Cond doesn't work unless it is captured completely with torch.compile",
@@ -3066,13 +3064,9 @@ def run_test_and_get_grads_loss(model, initial_hs, inputs):
         ).to(DEVICE)
 
         # Test 3 models: RNNScanList, RNNScanTensor, RNNLoop
-        models = [
-            ("ScanList", RNNScanList),
-            ("ScanTensor", RNNScanTensor),
-            ("Loop", RNNLoop),
-        ]
+        models = [RNNScanList, RNNScanTensor, RNNLoop]
 
-        for model_name, model_class in models:
+        for model_class in models:
             # Create uncompiled model
             model_uc = model_class().to(DEVICE)
             uncompiled_grads, uncompiled_loss = run_test_and_get_grads_loss(
@@ -7538,7 +7532,7 @@ def foo(x):
 
         inps = (torch.ones(3, 4), torch.ones(3, 5), torch.ones(5, 4), torch.ones(5, 3))
         for inp in inps:
-            gm = make_fx(foo, tracing_mode="symbolic")(torch.ones(3, 4))
+            gm = make_fx(foo, tracing_mode="symbolic")(inp)
             self.assertExpectedInline(
                 gm.code.strip(),
                 """\

test/inductor/test_flex_attention.py

@@ -5807,11 +5807,11 @@ def causal_mask(b, h, q_idx, kv_idx):
 
         from torch.utils._pytree import GetAttrKey
 
-        for key, tensor in tensors_with_keys:
+        for key, _tensor in tensors_with_keys:
             self.assertIsInstance(key, GetAttrKey)
             self.assertIsNotNone(key)
 
-        for key, value in context_with_keys:
+        for key, _value in context_with_keys:
             self.assertIsInstance(key, GetAttrKey)
             self.assertIsNotNone(key)
 

test/quantization/eager/test_bias_correction_eager.py

@@ -39,7 +39,7 @@ def correct_artificial_bias_quantize(self, float_model, img_data):
         torch.ao.quantization.convert(artificial_model, inplace=True)
 
         # manually changing bias
-        for name, submodule in artificial_model.named_modules():
+        for submodule in artificial_model.modules():
             if type(submodule) in _supported_modules:
                 x = get_param(submodule, "bias")
                 weight = get_param(submodule, "weight")

test/quantization/fx/test_quantize_fx.py

@@ -9663,10 +9663,10 @@ def forward(self, input: torch.Tensor, offsets: Optional[torch.Tensor] = None,
                 .set_global(get_default_qat_qconfig(qengine)) \
                 .set_object_type(torch.nn.EmbeddingBag, default_embedding_qat_qconfig)
 
-            train_indices = [[torch.randint(0, 10, (12, 12)), torch.randn((12, 1))] for _ in range(2)]
-            eval_output = [[torch.randint(0, 10, (12, 1))]]
+            train_indices = [[torch.randint(0, 10, (12, 12), device=device), torch.randn((12, 1), device=device)] for _ in range(2)]
+            eval_output = [[torch.randint(0, 10, (12, 1), device=device)]]
 
-            model = EmbeddingBagLinear().train()
+            model = EmbeddingBagLinear().to(device).train()
             prepared_fx_model = prepare_qat_fx(model, qconfig_dict, example_inputs=(train_indices[0][0],))
             test_only_train_fn(prepared_fx_model, train_indices)
             quant_model = convert_fx(prepared_fx_model,

test/test_nn.py

@@ -13516,7 +13516,7 @@ def compare_scaling(grads):
 
         # Should warning when parameters generator exhausted
         params = l.parameters()
-        for p in params:
+        for _p in params:
             pass
         with warnings.catch_warnings(record=True) as w:
             warnings.simplefilter("always")

test/test_transformers.py

@@ -2857,7 +2857,7 @@ def test_cudnn_attention_broken_166211(self):
         # https://github.com/pytorch/pytorch/issues/166211#issue-3551350377
         shape = (20, 4, 4, 32)
         scale = 10
-        for i in range(100):
+        for _ in range(100):
             q = torch.randn(*shape, device='cuda', dtype=torch.bfloat16) * scale
             k = torch.randn(*shape, device='cuda', dtype=torch.bfloat16) * scale
             v = torch.randn(*shape, device='cuda', dtype=torch.bfloat16) * scale