[2/N] Fix unused loop variables (pytorch#166500)

This PR removes unused loop variables.

Pull Request resolved: pytorch#166500
Approved by: https://github.com/mlazos

Author: cyyever

test/distributed/_composable/test_composability/test_2d_composability.py

@@ -218,7 +218,7 @@ def _shard_placement_fn(param: nn.Parameter) -> Optional[Shard]:
 
         torch.manual_seed(42 + global_mesh.get_local_rank("dp"))
         inp = torch.randint(0, model_args.vocab_size, (2, 16), device=device_type)
-        for iter_idx in range(5):
+        for _ in range(5):
             ref_loss = ref_model(inp).sum()
             loss = model(inp).sum()
             self.assertEqual(ref_loss, loss)
@@ -238,9 +238,7 @@ def _shard_placement_fn(param: nn.Parameter) -> Optional[Shard]:
             # runs its reduce-scatter
             self.assertIsInstance(model.pos_embeddings.weight.placements[1], Shard)
             self.assertIsInstance(model.pos_embeddings.weight.grad.placements[1], Shard)
-            for ref_param, (param_name, param) in zip(
-                ref_model.parameters(), model.named_parameters()
-            ):
+            for ref_param, param in zip(ref_model.parameters(), model.parameters()):
                 full_grad = param.grad.full_tensor()
                 self.assertEqual(ref_param.grad, full_grad)
 

test/distributed/_composable/test_composability/test_pp_composability.py

@@ -336,7 +336,7 @@ def apply_tp(
             for model in partial_models
         ]
 
-        for train_step in range(5):
+        for _train_step in range(5):
             for optimizer in optimizers:
                 optimizer.zero_grad()
             inputs = torch.rand((num_microbatches, dim), device=self.device)
@@ -517,7 +517,7 @@ def apply_same_precision(partial_model):
             for model in ref_partial_models
         ]
 
-        for train_step in range(5):
+        for _train_step in range(5):
             for optimizer in optimizers:
                 optimizer.zero_grad()
             for ref_optimizer in ref_optimizers:

test/quantization/core/test_quantized_op.py

@@ -2202,7 +2202,7 @@ def test_cat(self, X, num, dim, relu):
         X = torch.from_numpy(X)
         new_shape = np.array(X.shape)
         new_shape[dim] = 0
-        for idx in range(num):
+        for _ in range(num):
             tensors_q.append(torch.quantize_per_tensor(X, scale, zero_point,
                                                        torch_type))
             tensors_ref.append(X)
@@ -3162,7 +3162,7 @@ def forward(
                     # Quantize
                     mha_quantized = torch.ao.quantization.convert(mha_prepared)
 
-                    for name, param in mha_quantized.named_parameters():
+                    for name, _param in mha_quantized.named_parameters():
                         self.assertTrue("in_proj_weight" not in name)
 
                     qy = mha_quantized(*q_data)
@@ -5991,7 +5991,7 @@ def trace_handler(p):
                 activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
                 schedule=my_schedule,
                 on_trace_ready=trace_handler) as prof:
-            for i in range(30):
+            for _ in range(30):
                 conv_op(input, weight, None, stride, padding, dilation, groups)
                 prof.step()
 
@@ -6006,7 +6006,7 @@ def trace_handler(p):
                 activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
                 schedule=my_schedule,
                 on_trace_ready=trace_handler) as prof:
-            for i in range(30):
+            for _ in range(30):
                 conv_op(input_fp16, weight_fp16, None, stride, padding, dilation, groups)
                 prof.step()
 
@@ -6023,7 +6023,7 @@ def trace_handler(p):
                 activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
                 schedule=my_schedule,
                 on_trace_ready=trace_handler) as prof:
-            for i in range(30):
+            for _ in range(30):
                 conv_op(input_int8, weight_prepacked, scale, zero_point)
                 prof.step()
 

test/quantization/core/test_quantized_tensor.py

@@ -586,7 +586,7 @@ def test_compare_per_channel_device_numerics(self):
         ]
         axis = 1
         device = torch.device('cuda')
-        for i in range(20):
+        for _ in range(20):
             for dtype, zero_type in dtype_and_zero_types:
                 r = torch.rand(2, 2) * 10
                 r[0, 0] = 2.5
@@ -1113,7 +1113,7 @@ def _test_qtensor_index_put_non_accumulate_deterministic(self, device):
             zero_point = 10
             types = [torch.qint8, torch.quint8, torch.qint32]
             for qtype in types:
-                for i in range(3):
+                for _ in range(3):
                     m = random.randint(10, 20)
                     elems = random.randint(20000, 30000)
                     values = torch.rand(elems, device=device)

test/quantization/core/test_workflow_module.py

@@ -1028,7 +1028,7 @@ def test_observers_preserve_buffers(self):
         for observer_type in observer_types:
             observer = observer_type()
             buffer_ids_before = _get_buffer_ids(observer)
-            for _i in range(5):
+            for _ in range(5):
                 inputs = torch.rand((4, 4, 4))
                 observer(inputs)
             buffer_ids_after = _get_buffer_ids(observer)
@@ -1046,7 +1046,7 @@ def test_fake_quant_preserves_buffers(self):
         """
         model = torch.ao.quantization.FakeQuantize()
         buffer_ids_before = _get_buffer_ids(model)
-        for _i in range(5):
+        for _ in range(5):
             inputs = torch.rand((4, 4, 4))
             model(inputs)
         model.apply(torch.ao.quantization.enable_fake_quant)
@@ -1309,7 +1309,7 @@ def test_compare_fused_obs_fq_oss_module(self, device):
         torch.ao.quantization.enable_observer(mod_ref)
         mod_ref.to(device)
 
-        for i in range(10):
+        for _ in range(10):
             x = torch.randn(5, 5, device=device)
             out = mod(x)
             out_ref = mod_ref(x)
@@ -1445,7 +1445,7 @@ def test_embedding_qat_config(self):
 
                 count_fake_quant = 0
                 count_activation_postproc = 0
-                for name, mod in quant_model.named_modules():
+                for name, _mod in quant_model.named_modules():
                     if name.endswith('weight_fake_quant'):
                         count_fake_quant += 1
                     if name.count('activation_post_process') == 1 and 'weight_fake_quant' not in name:

test/torch_np/numpy_tests/core/test_multiarray.py

@@ -2702,7 +2702,7 @@ def test_diagonal_memleak(self):
         a = np.zeros((100, 100))
         if HAS_REFCOUNT:
             assert_(sys.getrefcount(a) < 50)
-        for i in range(100):
+        for _ in range(100):
             a.diagonal()
         if HAS_REFCOUNT:
             assert_(sys.getrefcount(a) < 50)
@@ -4457,7 +4457,7 @@ def test_none_shape(self):
 
     def test_0d_shape(self):
         # to it multiple times to test it does not break alloc cache gh-9216
-        for i in range(10):
+        for _ in range(10):
             x = np.empty((1,))
             x.resize(())
             assert_equal(x.shape, ())
@@ -5081,7 +5081,7 @@ def test_dot_3args(self):
         v = np.random.random_sample((16, 32))
 
         r = np.empty((1024, 32))
-        for i in range(12):
+        for _ in range(12):
             dot(f, v, r)
         if HAS_REFCOUNT:
             assert_equal(sys.getrefcount(r), 2)
@@ -6678,7 +6678,7 @@ def test_largedim(self):
         np.random.seed(2)
         array = np.random.rand(*shape)
 
-        for i in range(10):
+        for _ in range(10):
             benchmark = array.nonzero()
             result = array.nonzero()
             assert_array_equal(benchmark, result)

test/torch_np/numpy_tests/fft/test_pocketfft.py

@@ -366,7 +366,7 @@ def worker(args, q):
 
         [x.join() for x in t]
         # Make sure all threads returned the correct value
-        for i in range(self.threads):
+        for _ in range(self.threads):
             # under torch.dynamo `assert_array_equal` fails with relative errors of
             # about 1.5e-14. Hence replace it with `assert_allclose(..., rtol=2e-14)`
             assert_allclose(

test/torch_np/numpy_tests/lib/test_function_base.py

@@ -2596,12 +2596,12 @@ def test_dtype_reference_leaks(self):
         intp_refcount = sys.getrefcount(np.dtype(np.intp))
         double_refcount = sys.getrefcount(np.dtype(np.double))
 
-        for j in range(10):
+        for _ in range(10):
             np.bincount([1, 2, 3])
         assert_equal(sys.getrefcount(np.dtype(np.intp)), intp_refcount)
         assert_equal(sys.getrefcount(np.dtype(np.double)), double_refcount)
 
-        for j in range(10):
+        for _ in range(10):
             np.bincount([1, 2, 3], [4, 5, 6])
         assert_equal(sys.getrefcount(np.dtype(np.intp)), intp_refcount)
         assert_equal(sys.getrefcount(np.dtype(np.double)), double_refcount)

test/torch_np/numpy_tests/linalg/test_linalg.py

@@ -1706,7 +1706,7 @@ def test_reduced_rank(self):
         # Test matrices with reduced rank
         #  rng = np.random.RandomState(20120714)
         np.random.seed(20120714)
-        for i in range(100):
+        for _ in range(100):
             # Make a rank deficient matrix
             X = np.random.normal(size=(40, 10))
             X[:, 0] = X[:, 1] + X[:, 2]

torch/_dynamo/compiled_autograd.py

@@ -969,7 +969,8 @@ def dce(self) -> None:
 
         # Dynamo guards will error instead of creating aliasing guards unless we unpack them in the graph
         unpack_nodes: OrderedSet[torch.fx.Node] = OrderedSet()
-        for i, node in enumerate(self.fx_tracer.graph.find_nodes(op="placeholder")):
+        i: int | None = None
+        for i, node in enumerate(self.fx_tracer.graph.find_nodes(op="placeholder")):  # noqa: B007
             unpack_nodes.update(node.users.keys())
         assert i == len(_graph_placeholders) - 1