Fix unused Python variables outside torch/ and test/ (pytorch#136359)

Pull Request resolved: pytorch#136359
Approved by: https://github.com/albanD

Author: rec

.github/scripts/filter_test_configs.py

@@ -332,7 +332,7 @@ def process_jobs(
         # The job name from github is in the PLATFORM / JOB (CONFIG) format, so breaking
         # it into its two components first
         current_platform, _ = (n.strip() for n in job_name.split(JOB_NAME_SEP, 1) if n)
-    except ValueError as error:
+    except ValueError:
         warnings.warn(f"Invalid job name {job_name}, returning")
         return test_matrix
 

.github/scripts/runner_determinator.py

@@ -258,7 +258,7 @@ def load_yaml(yaml_text: str) -> Any:
     try:
         data = yaml.safe_load(yaml_text)
         return data
-    except yaml.YAMLError as exc:
+    except yaml.YAMLError:
         log.exception("Error loading YAML")
         raise
 

.github/scripts/test_trymerge.py

@@ -898,7 +898,7 @@ def test_dont_ignore_flaky_failures(self, *args: Any) -> None:
         repo = DummyGitRepo()
         # Check that failure is classified as flaky but still raises exception
         with warnings.catch_warnings(record=True) as w, self.assertRaises(RuntimeError):
-            rule = find_matching_merge_rule(pr, repo)
+            find_matching_merge_rule(pr, repo)
         self.assertEqual(len(w), 1)
         self.assertIn(
             "1 checks failed but were likely due flakiness or broken trunk",

.github/scripts/trymerge.py

@@ -1747,7 +1747,7 @@ def get_readable_drci_results(drci_classifications: Any) -> str:
         try:
             print(f"From Dr.CI checkrun summary: {drci_summary}")
             drci_classifications = json.loads(str(drci_summary))
-        except json.JSONDecodeError as error:
+        except json.JSONDecodeError:
             warn("Invalid Dr.CI checkrun summary")
             drci_classifications = {}
 
@@ -1918,7 +1918,6 @@ def do_revert_prs(
     dry_run: bool = False,
 ) -> None:
     # Prepare and push revert commits
-    commit_shas: List[str] = []
     for commit_sha, pr in shas_and_prs:
         revert_msg = f"\nReverted {pr.get_pr_url()} on behalf of {prefix_with_github_url(author_login)}"
         revert_msg += extra_msg

.github/workflows/_runner-determinator.yml

@@ -326,7 +326,7 @@ jobs:
               try:
                   data = yaml.safe_load(yaml_text)
                   return data
-              except yaml.YAMLError as exc:
+              except yaml.YAMLError:
                   log.exception("Error loading YAML")
                   raise
 

benchmarks/distributed/rpc/rl/coordinator.py

@@ -72,7 +72,6 @@ def run_coordinator(self, episodes, episode_steps, queue):
             print(f"Episode {ep} - ", end="")
 
             n_steps = episode_steps
-            agent_start_time = time.time()
 
             futs = []
             for ob_rref in self.ob_rrefs:

benchmarks/dynamo/check_accuracy.py

@@ -19,7 +19,7 @@
 def get_field(csv, model_name: str, field: str):
     try:
         return csv.loc[csv["name"] == model_name][field].item()
-    except Exception as e:
+    except Exception:
         return None
 
 

benchmarks/dynamo/check_graph_breaks.py

@@ -9,7 +9,7 @@
 def get_field(csv, model_name: str, field: str):
     try:
         return csv.loc[csv["name"] == model_name][field].item()
-    except Exception as e:
+    except Exception:
         return None
 
 

benchmarks/dynamo/common.py

@@ -671,7 +671,7 @@ def print_summary_table(data, print_dataframe=False):
                     col.ljust(width),
                     f"gmean={gmean(cdata):.2f}x mean={cdata.mean():.3f}x",
                 )
-        except Exception as e:
+        except Exception:
             pass
 
 
@@ -3018,7 +3018,7 @@ def record_status(accuracy_status, dynamo_start_stats):
                     )
                 ):
                     is_same = False
-            except Exception as e:
+            except Exception:
                 # Sometimes torch.allclose may throw RuntimeError
                 is_same = False
 
@@ -3110,7 +3110,7 @@ def record_status(accuracy_status, dynamo_start_stats):
                     tol=tolerance,
                 ):
                     is_same = False
-            except Exception as e:
+            except Exception:
                 # Sometimes torch.allclose may throw RuntimeError
                 is_same = False
 
@@ -3157,7 +3157,7 @@ def check_tolerance(
                 self.init_optimizer(name, current_device, model.parameters())
                 optimized_model_iter_fn = optimize_ctx(self.run_n_iterations)
                 new_result = optimized_model_iter_fn(model, example_inputs)
-            except Exception as e:
+            except Exception:
                 log.exception("")
                 print(
                     "TorchDynamo optimized model failed to run because of following error"
@@ -3542,7 +3542,7 @@ def minify_model(
 
         try:
             shutil.move("repro.py", f"{repro_dir}/{name}_repro.py")
-        except OSError as e:
+        except OSError:
             logging.error("Could not find repro script for model %s", name)
         else:
             logging.info(
@@ -4369,9 +4369,6 @@ def run(runner, args, original_dir=None):
             # Set translation validation on by default on CI accuracy runs.
             torch.fx.experimental._config.translation_validation = True
 
-        ci = functools.partial(
-            CI, args.backend, training=args.training, dynamic=args.dynamic_shapes
-        )
     if args.ddp:
         assert args.training, "DDP benchmark requires --training mode"
         torch._dynamo.config.optimize_ddp = args.optimize_ddp_mode

benchmarks/dynamo/dist_util.py

@@ -90,7 +90,7 @@ def model_iter_fn(model, example_inputs, collect_outputs=False):
 
 def get_model(args):
     if args.torchbench_model:
-        old_cwd = setup_torchbench_cwd()
+        setup_torchbench_cwd()
         module = importlib.import_module(
             f"torchbenchmark.models.{args.torchbench_model}"
         )

benchmarks/dynamo/runner.py

@@ -1451,7 +1451,7 @@ def update(self):
             RegressionDetector(self.args).generate_comment()
             try:
                 RegressionTracker(self.args).diff()
-            except Exception as e:
+            except Exception:
                 logging.exception("")
                 with open(f"{self.args.output_dir}/gh_regression.txt", "w") as gh_fh:
                     gh_fh.write("")

benchmarks/dynamo/torchbench.py

@@ -236,7 +236,6 @@ def load_model(
             )
         is_training = self.args.training
         use_eval_mode = self.args.use_eval_mode
-        dynamic_shapes = self.args.dynamic_shapes
         candidates = [
             f"torchbenchmark.models.{model_name}",
             f"torchbenchmark.canary_models.{model_name}",

benchmarks/fastrnns/bench.py

@@ -205,7 +205,7 @@ def bench(rnn_runners, group_name, print_json=False, sep=" ", **params):
                 result_with_no_info = result._replace(info_fwd="None", info_bwd="None")
                 print_stderr(pretty_print(result_with_no_info, sep=sep))
                 results[name] = result
-            except Exception as e:
+            except Exception:
                 if not print_json:
                     raise
 

benchmarks/fastrnns/factory.py

@@ -338,8 +338,8 @@ def forward(input, hidden):
         seq_len = len(input.unbind(0))
         hy, cy = new_hidden
         for i in range(seq_len):
-            ln_i_output = ln_i(ln_input1)
-            ln_h_output = ln_h(ln_input1)
+            ln_i(ln_input1)
+            ln_h(ln_input1)
             cy = ln_c(cy)
 
         return out, (hy, cy)

benchmarks/fastrnns/test_bench.py

@@ -40,7 +40,7 @@ def cuda_sync(func, *args, **kwargs):
 class TestBenchNetwork:
     # See 'modeldef' fixture, which provides the things to benchmark
     def test_forward(self, modeldef, benchmark):
-        forward_output = benchmark(cuda_sync, modeldef.forward, *modeldef.inputs)
+        benchmark(cuda_sync, modeldef.forward, *modeldef.inputs)
 
     def test_backward(self, modeldef, benchmark):
         backward_input = modeldef.forward(*modeldef.inputs)

benchmarks/framework_overhead_benchmark/framework_overhead_benchmark.py

@@ -25,7 +25,7 @@
 
 
 def parse_op_args(op):
-    op_list = op.split(",")
+    op_list = op.split(",")  # noqa: F841
 
 
 def print_results(result):

benchmarks/functional_autograd_benchmark/functional_autograd_benchmark.py

@@ -190,9 +190,9 @@ def run_once(model: Callable, inp: InputsType, task: str, v: VType, **kwargs) ->
     func = get_task_func(task)
 
     if v is not None:
-        res = func(model, inp, v=v, strict=True)
+        func(model, inp, v=v, strict=True)
     else:
-        res = func(model, inp, strict=True)
+        func(model, inp, strict=True)
 
 
 def run_once_functorch(

benchmarks/fuser/run_benchmarks.py

@@ -284,7 +284,6 @@ def run_benchmarks(operators, shapes):
         shapes = [globals()[k] for k in shapes.split(",")]
 
     print("fuser,device,operator,shape,time")
-    results = []
     for shape, operator in itertools.product(shapes, operators):
         nargs = len(inspect.signature(operator).parameters)
         args = shape()

benchmarks/gpt_fast/mixtral_moe_quantize.py

@@ -132,7 +132,6 @@ def __init__(
         target_dtype=None,
     ) -> None:
         assert target_dtype is not None
-        factory_kwargs = {"device": device, "dtype": dtype}
         super().__init__()
         self.in_features = in_features
         self.out_features = out_features

benchmarks/gpt_fast/quantize.py

@@ -93,7 +93,6 @@ def __init__(
         device=None,
         dtype=None,
     ) -> None:
-        factory_kwargs = {"device": device, "dtype": dtype}
         super().__init__()
         self.in_features = in_features
         self.out_features = out_features

benchmarks/instruction_counts/core/utils.py

@@ -74,7 +74,7 @@ def parse_stmts(stmts: str) -> Tuple[str, str]:
     assert len(lines) >= 3, f"Invalid string:\n{stmts}"
 
     column_header_pattern = r"^Python\s{35}\| C\+\+(\s*)$"
-    signature_pattern = r"^: f\((.*)\)( -> (.+))?\s*$"
+    signature_pattern = r"^: f\((.*)\)( -> (.+))?\s*$"  # noqa: F841
     separation_pattern = r"^[-]{40} | [-]{40}$"
     code_pattern = r"^(.{40}) \|($| (.*)$)"
 

benchmarks/nested/nested_bmm_bench.py

@@ -6,14 +6,14 @@
 
 def bench(nt_a, nt_b, niter):
     # Warmup
-    nt_c = nt_a.bmm(nt_b)
+    nt_a.bmm(nt_b)
 
     torch.cuda.synchronize()
     start_event = torch.cuda.Event(enable_timing=True)
     end_event = torch.cuda.Event(enable_timing=True)
     start_event.record()
     for iter in range(niter):
-        nt_c = nt_a.bmm(nt_b)
+        nt_a.bmm(nt_b)
     end_event.record()
     torch.cuda.synchronize()
     runtime = (start_event.elapsed_time(end_event)) / niter

benchmarks/operator_benchmark/benchmark_core.py

@@ -111,10 +111,9 @@ def _build_test(
 
         if tags is None:
             raise ValueError("Missing tags in configs")
-        input_config = str(test_attrs)[1:-1].replace("'", "")
+
         op = bench_op()
         assert op is not None, "Can't create test"
-        tensor_error_info = None
         # op_name_function is a dictionary which has op_name and op_function.
         # an example of op_name_function is:
         # {'op_name' : 'abs', 'op_function' : torch.abs}

benchmarks/operator_benchmark/pt/qrnn_test.py

@@ -31,8 +31,6 @@ def init(self, I, H, NL, B, D, dtype):
 
         # The quantized.dynamic.LSTM has a bug. That's why we create a regular
         # LSTM, and quantize it later. See issue #31192.
-        scale = 1.0 / 256
-        zero_point = 0
         cell_nn = nn.LSTM(
             input_size=I,
             hidden_size=H,

benchmarks/profiler_benchmark/profiler_bench.py

@@ -97,7 +97,7 @@ def payload():
                     with_stack=args.with_stack,
                     use_kineto=args.use_kineto,
                     use_cpu=not args.cuda_only,
-                ) as prof:
+                ):
                     x = workload(input_x)
                 return x
 

benchmarks/serialization/simple_measurement.py

@@ -13,14 +13,14 @@ def benchmark(self):
             torch.save(x, "big_tensor.zip", _use_new_zipfile_serialization=use_new)
 
         with Timer() as big2:
-            v = torch.load("big_tensor.zip")
+            torch.load("big_tensor.zip")
 
         x = [torch.ones(10, 10) for i in range(200)]
         with Timer() as small1:
             torch.save(x, "small_tensor.zip", _use_new_zipfile_serialization=use_new)
 
         with Timer() as small2:
-            v = torch.load("small_tensor.zip")
+            torch.load("small_tensor.zip")
 
         return {
             "Big Tensors Save": big1.ms_duration,

benchmarks/sparse/dlmc/utils.py

@@ -56,16 +56,12 @@ def load_sparse_matrix(path, device):
 def gen_vector(path, device):
     with open(path) as file:
         nrows, ncols, nnz = (int(el) for el in file.readline().split(", "))
-        index_pointers = (int(el) for el in file.readline().split())
-        indices = (int(el) for el in file.readline().split())
         return torch.randn(nrows, dtype=torch.double, device=device)
 
 
 def gen_matrix(path, device):
     with open(path) as file:
         nrows, ncols, nnz = (int(el) for el in file.readline().split(", "))
-        index_pointers = (int(el) for el in file.readline().split())
-        indices = (int(el) for el in file.readline().split())
         return torch.randn(nrows, ncols, dtype=torch.double, device=device)
 
 

benchmarks/sparse/triton_ops.py

@@ -374,7 +374,7 @@ def show_best_messages(best_messages=best_messages):
                 for r in range(args.repeat):
                     try:
                         time_ms, performance_tflops = test_func(x, y, **meta)
-                    except triton.compiler.OutOfResources as msg:
+                    except triton.compiler.OutOfResources:
                         print(
                             f"op={op}[{meta_str}]({bsr_size},{k}x{n}) dtype={args.dtype} {sparsity=}(nnz={x._nnz()})"
                             f" blocksize={bm}x{bk} OutOfResources",

benchmarks/tensorexpr/elementwise.py

@@ -208,7 +208,6 @@ def module(cls):
         return "simple_element"
 
     def memory_workload(self):
-        input_count = len(self.inputs)
         if self.mode == "fwd":
             sol_count = 2
             algorithmic_count = 2

benchmarks/transformer/better_transformer_vs_mha_functional.py

@@ -136,7 +136,7 @@ def run(
         torch.testing.assert_close(
             y_native_mha_fast, y_native_mha_slow, atol=1e-3, rtol=1e-3
         )
-    except AssertionError as e:
+    except AssertionError:
         error_dict[entry_name] += 1
         pprint(error_dict)
 

benchmarks/transformer/score_mod.py

@@ -98,8 +98,6 @@ def generate_inputs(
 
     assert q_heads % kv_heads == 0
 
-    num_h_groups = q_heads // kv_heads
-
     make_q = partial(
         torch.rand, q_shape, device=device, dtype=dtype, requires_grad=requires_grad
     )

benchmarks/transformer/sdp.py

@@ -211,7 +211,7 @@ def run_single_experiment(config: ExperimentConfig) -> ExperimentResults:
         enable_flash=config.enable_flash,
         enable_mem_efficient=config.enable_mem_efficient,
         enable_cudnn=config.enable_cudnn,
-    ) as kernel_choice, torch.inference_mode() as inference_mode:
+    ):
         dropout_p = 0.0
         mask = None
 

functorch/dim/delayed_mul_tensor.py

@@ -62,7 +62,6 @@ def to_char(d):
 
         plhs, levelslhs = self._lhs._tensor, self._lhs._levels
         prhs, levelsrhs = self._rhs._tensor, self._rhs._levels
-        new_dims = tuple(d for d in self.dims if d not in dims)
         new_levels = [l for l in self._levels if l not in dims]
         fmt = "".join(
             [

functorch/dim/reference.py

@@ -198,7 +198,6 @@ def unwrap(t):
 
     if orig in pointwise:
         result_levels = llist()
-        arg_levels = llist()
         to_expand = []
         for i, f in enumerate(flat_args):
             if isinstance(f, TensorLike):
@@ -268,7 +267,6 @@ def positional(self, *dims):
             needs_view = True
 
     permute = list(range(len(levels)))
-    nflat = len(flat_dims)
     for i, d in enumerate(flat_dims):
         try:
             idx = levels.index(d)