[BE] fix ruff rule E226: add missing whitespace around operator in f-strings (#144415)

The fixes are generated by:

```bash
ruff check --fix --preview --unsafe-fixes --select=E226 .
lintrunner -a --take "RUFF,PYFMT" --all-files
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/144415
Approved by: https://github.com/huydhn, https://github.com/Skylion007

.github/scripts/trymerge.py

@@ -669,7 +669,7 @@ def get_ghstack_prs(
         if not open_only or not candidate.is_closed():
             return False
         print(
-            f"Skipping {idx+1} of {len(rev_list)} PR (#{candidate.pr_num}) as its already been merged"
+            f"Skipping {idx + 1} of {len(rev_list)} PR (#{candidate.pr_num}) as its already been merged"
         )
         return True
 

benchmarks/dynamo/common.py

@@ -676,7 +676,7 @@ def print_summary_table(data, print_dataframe=False):
                 print(col.ljust(width), f"mean={data[col].mean():.3f}x")
             elif col in ("accuracy"):
                 pass_rate = (data[col] == "pass").mean()
-                print(col.ljust(width), f"pass_rate={100*pass_rate:.2f}%")
+                print(col.ljust(width), f"pass_rate={100 * pass_rate:.2f}%")
             else:
                 cdata = data[col]
                 print(
@@ -4993,7 +4993,7 @@ def run(runner, args, original_dir=None):
         for i, name in enumerate(model_names):
             current_name = name
             if args.progress:
-                print(f"Running model {i+1}/{nmodels}", flush=True)
+                print(f"Running model {i + 1}/{nmodels}", flush=True)
 
             try:
                 timeout = args.timeout

benchmarks/dynamo/microbenchmarks/cache_debug_microbenchmarks.py

@@ -25,7 +25,7 @@ def main():
         return details.debug_lines()
 
     t = min(timeit.repeat(fn, number=K, repeat=3))
-    print(f"iterating over {N*K} FX nodes took {t:.1f}s ({N*K/t:.0f} nodes/s)")
+    print(f"iterating over {N * K} FX nodes took {t:.1f}s ({N * K / t:.0f} nodes/s)")
 
 
 if __name__ == "__main__":

benchmarks/dynamo/microbenchmarks/fx_microbenchmarks.py

@@ -24,7 +24,7 @@ def main():
             pass
 
     t = min(timeit.repeat(fn, number=K, repeat=3))
-    print(f"iterating over {N*K} FX nodes took {t:.1f}s ({N*K/t:.0f} nodes/s)")
+    print(f"iterating over {N * K} FX nodes took {t:.1f}s ({N * K / t:.0f} nodes/s)")
 
 
 if __name__ == "__main__":

benchmarks/dynamo/microbenchmarks/overheads.py

@@ -19,7 +19,7 @@ def bench(name, fn, requires_grad):
     end = time.perf_counter()
 
     results = timeit.repeat(lambda: fn(x), number=1000, repeat=1000)
-    print(f"{name} {np.median(results)*1000:.1f}us (warmup={end-start:.1f}s)")
+    print(f"{name} {np.median(results) * 1000:.1f}us (warmup={end - start:.1f}s)")
 
 
 def main():

benchmarks/dynamo/pr_time_benchmarks/check_results.py

@@ -144,7 +144,7 @@ def main():
             fail = True
             print(
                 f"REGRESSION: benchmark {key} failed, actual result {result} "
-                f"is {ratio:.2f}% higher than expected {entry.expected_value} ±{entry.noise_margin*100:+.2f}% "
+                f"is {ratio:.2f}% higher than expected {entry.expected_value} ±{entry.noise_margin * 100:+.2f}% "
                 f"if this is an expected regression, please update the expected results.\n"
             )
             print(
@@ -158,7 +158,7 @@ def main():
 
             print(
                 f"WIN: benchmark {key} failed, actual result {result} is {ratio:+.2f}% lower than "
-                f"expected {entry.expected_value} ±{entry.noise_margin*100:.2f}% "
+                f"expected {entry.expected_value} ±{entry.noise_margin * 100:.2f}% "
                 f"please update the expected results. \n"
             )
             print(
@@ -170,7 +170,7 @@ def main():
         else:
             print(
                 f"PASS: benchmark {key} pass, actual result {result} {ratio:+.2f}% is within "
-                f"expected {entry.expected_value} ±{entry.noise_margin*100:.2f}%\n"
+                f"expected {entry.expected_value} ±{entry.noise_margin * 100:.2f}%\n"
             )
 
             log("pass")

benchmarks/dynamo/runner.py

@@ -543,7 +543,7 @@ def build_summary(args):
         out_io.write(f"Number CUDA Devices: {torch.cuda.device_count()}\n")
         out_io.write(f"Device Name: {torch.cuda.get_device_name(0)}\n")
         out_io.write(
-            f"Device Memory [GB]: {torch.cuda.get_device_properties(0).total_memory/1e9}\n"
+            f"Device Memory [GB]: {torch.cuda.get_device_properties(0).total_memory / 1e9}\n"
         )
 
     title = "## Build Summary"

benchmarks/dynamo/training_loss.py

@@ -193,9 +193,9 @@ def main():
     print(
         f"Train model on {args.epochs} epochs with backend {args.backend} and optimizer {args.optimizer}:"
     )
-    print(f"PyTorch spent {timedelta(seconds=native_elapsed/args.epochs)} per epoch")
+    print(f"PyTorch spent {timedelta(seconds=native_elapsed / args.epochs)} per epoch")
     print(
-        f"TorchDynamo spent {timedelta(seconds=dynamo_elapsed/args.epochs)} per epoch"
+        f"TorchDynamo spent {timedelta(seconds=dynamo_elapsed / args.epochs)} per epoch"
     )
 
 

functorch/examples/compilation/fuse_module.py

@@ -51,6 +51,6 @@ for a, b in zip(run(mod, input), run(compiled_mod, input)):
 for _ in range(5):
     i = 10000
     t = timeit.Timer("mod(input)", globals=globals()).timeit(10000)
-    print(f"eager {t/i*1e6}")
+    print(f"eager {t / i * 1e6}")
     t = timeit.Timer("compiled_mod(input)", globals=globals()).timeit(10000)
-    print(f"compiled {t/i*1e6}")
+    print(f"compiled {t / i * 1e6}")

functorch/examples/maml_omniglot/maml-omniglot-higher.py

@@ -235,7 +235,7 @@ def test(db, net, device, epoch, log):
 
     qry_losses = torch.cat(qry_losses).mean().item()
     qry_accs = 100.0 * torch.cat(qry_accs).float().mean().item()
-    print(f"[Epoch {epoch+1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
+    print(f"[Epoch {epoch + 1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
     log.append(
         {
             "epoch": epoch + 1,

functorch/examples/maml_omniglot/maml-omniglot-ptonly.py

@@ -225,7 +225,7 @@ def test(db, net, device, epoch, log):
 
     qry_losses = torch.cat(qry_losses).mean().item()
     qry_accs = 100.0 * torch.cat(qry_accs).float().mean().item()
-    print(f"[Epoch {epoch+1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
+    print(f"[Epoch {epoch + 1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
     log.append(
         {
             "epoch": epoch + 1,

functorch/examples/maml_omniglot/maml-omniglot-transforms.py

@@ -225,7 +225,7 @@ def test(db, net, device, epoch, log):
 
     qry_losses = torch.cat(qry_losses).mean().item()
     qry_accs = 100.0 * torch.cat(qry_accs).float().mean().item()
-    print(f"[Epoch {epoch+1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
+    print(f"[Epoch {epoch + 1:.2f}] Test Loss: {qry_losses:.2f} | Acc: {qry_accs:.2f}")
     log.append(
         {
             "epoch": epoch + 1,

test/export/random_dag.py

@@ -137,7 +137,7 @@ class Unflatten(TestGenerator):
                 code = Block()
                 code.new_line("super().__init__()")
                 if i < self.n - 1:
-                    code.new_line(f"self.n{i+1} = N{i+1}()")
+                    code.new_line(f"self.n{i + 1} = N{i + 1}()")
                 return code
 
             def gen_forward_body(self, i: int):
@@ -207,7 +207,7 @@ class ConstantUnflatten(Unflatten):
                 code.new_line("super().__init__()")
                 code.new_line("self.const = torch.ones(1)")
                 if i < self.n - 1:
-                    code.new_line(f"self.n{i+1} = N{i+1}()")
+                    code.new_line(f"self.n{i + 1} = N{i + 1}()")
                 return code
 
             def gen_forward_body(self, i: int):
@@ -249,7 +249,7 @@ class BufferUnflatten(Unflatten):
                 code.new_line("super().__init__()")
                 code.new_line("self.buf = torch.nn.Buffer(torch.ones(1))")
                 if i < self.n - 1:
-                    code.new_line(f"self.n{i+1} = N{i+1}()")
+                    code.new_line(f"self.n{i + 1} = N{i + 1}()")
                 return code
 
             def gen_forward_body(self, i: int):

test/test_testing.py

@@ -489,7 +489,7 @@ if __name__ == '__main__':
         del env[PYTORCH_TESTING_DEVICE_ONLY_FOR_KEY]
         env[PYTORCH_TESTING_DEVICE_EXCEPT_FOR_KEY] = 'cpu'
         _, stderr = TestCase.run_process_no_exception(test_filter_file_template, env=env)
-        self.assertIn(f'Ran {test_bases_count-1} test', stderr.decode('ascii'))
+        self.assertIn(f'Ran {test_bases_count - 1} test', stderr.decode('ascii'))
 
         # Test with setting both should throw exception
         env[PYTORCH_TESTING_DEVICE_ONLY_FOR_KEY] = 'cpu'

third_party/generate-xnnpack-wrappers.py

@@ -100,7 +100,7 @@ IGNORED_SOURCES = set((
 def handle_singleline_parse(line):
     start_index = line.find("(")
     end_index = line.find(")")
-    line = line[start_index+1:end_index]
+    line = line[start_index + 1:end_index]
     key_val = line.split(" ")
     return key_val[0], [x[4:] for x in key_val[1:]]
 

torch/_dynamo/backends/debugging.py

@@ -312,7 +312,7 @@ class ExplainOutput:
 
         output += "Break Reasons:\n"
         for idx, break_reason in enumerate(self.break_reasons):
-            output += f"  Break Reason {idx+1}:\n"
+            output += f"  Break Reason {idx + 1}:\n"
             output += f"    Reason: {break_reason.reason}\n"
             output += "    User Stack:\n"
             for frame_summary in break_reason.user_stack:
@@ -321,14 +321,14 @@ class ExplainOutput:
         if self.ops_per_graph is not None:
             output += "Ops per Graph:\n"
             for idx, ops in enumerate(self.ops_per_graph):
-                output += f"  Ops {idx+1}:\n"
+                output += f"  Ops {idx + 1}:\n"
                 for op in ops:
                     output += f"    {op}\n"
 
         if self.out_guards is not None:
             output += "Out Guards:\n"
             for i, guard in enumerate(self.out_guards):
-                output += f"  Guard {i+1}:\n"
+                output += f"  Guard {i + 1}:\n"
                 output += f"    {str(guard)}"
 
         if self.compile_times is not None:

torch/_dynamo/debug_utils.py

@@ -184,7 +184,7 @@ class NNModuleToString:
             example_param = next(module.parameters(), None)
             if example_param is not None and example_param.is_cuda:
                 module_str = f"{module_str}.cuda()"
-            model_str += f"{tab*2}self.{module_name} = {module_str}\n"
+            model_str += f"{tab * 2}self.{module_name} = {module_str}\n"
 
         for buffer_name, buffer in gm._buffers.items():
             if buffer is None:
@@ -203,7 +203,9 @@ class NNModuleToString:
                 )
             if buffer.is_cuda:
                 tensor_str = f"{tensor_str}.cuda()"
-            model_str += f"{tab*2}self.register_buffer('{buffer_name}', {tensor_str})\n"
+            model_str += (
+                f"{tab * 2}self.register_buffer('{buffer_name}', {tensor_str})\n"
+            )
 
         for param_name, param in gm._parameters.items():
             if param is None:
@@ -212,7 +214,7 @@ class NNModuleToString:
             if param.is_cuda:
                 maybe_device = ', device="cuda"'
             tensor_str = f"torch.nn.Parameter(torch.randn({list(param.shape)}, dtype={param.dtype}{maybe_device}))"
-            model_str += f"{tab*2}self.{param_name} = {tensor_str}\n"
+            model_str += f"{tab * 2}self.{param_name} = {tensor_str}\n"
 
         # TODO - Keep this code for now. But, I don't think we will need this.
         # attrs = dir(gm)

torch/_inductor/wrapper_benchmark.py

@@ -396,7 +396,7 @@ def compiled_module_main(benchmark_name, benchmark_compiled_module_fn):
 
         if torch.cuda.is_available():
             peak_mem = torch.cuda.max_memory_allocated()
-            print(f"Peak GPU memory usage {peak_mem/1e6:.3f} MB")
+            print(f"Peak GPU memory usage {peak_mem / 1e6:.3f} MB")
 
         if torch.cuda.is_available() and args.cuda_memory_snapshot:
             collect_memory_snapshot(benchmark_compiled_module_fn)

torch/_logging/_internal.py

@@ -844,7 +844,7 @@ class TorchLogsFormatter(logging.Formatter):
         filepath = make_module_path_relative(record.pathname)
 
         prefix = (
-            f"{record.rankprefix}{shortlevel}{record.asctime}.{int(record.msecs*1000):06d} {record.process} "
+            f"{record.rankprefix}{shortlevel}{record.asctime}.{int(record.msecs * 1000):06d} {record.process} "
             f"{filepath}:"
             f"{record.lineno}]{record.traceid}{record.artifactprefix}"
         )

torch/_numpy/testing/utils.py

@@ -2274,7 +2274,7 @@ def check_free_memory(free_bytes):
             )
 
         msg = (
-            f"{free_bytes/1e9} GB memory required, but environment variable "
+            f"{free_bytes / 1e9} GB memory required, but environment variable "
             f"NPY_AVAILABLE_MEM={env_value} set"
         )
     else:
@@ -2288,9 +2288,7 @@ def check_free_memory(free_bytes):
             )
             mem_free = -1
         else:
-            msg = (
-                f"{free_bytes/1e9} GB memory required, but {mem_free/1e9} GB available"
-            )
+            msg = f"{free_bytes / 1e9} GB memory required, but {mem_free / 1e9} GB available"
 
     return msg if mem_free < free_bytes else None
 

torch/distributed/_tools/ilp_utils.py

@@ -257,11 +257,11 @@ def display_bytes(b: int, unit: str = "MiB") -> str:
     return a string that represent the number of bytes in a desired unit
     """
     if unit == "KiB":
-        return f"{b/2**10:.2f} KiB"
+        return f"{b / 2**10:.2f} KiB"
     if unit == "MiB":
-        return f"{b/2**20:.2f} MiB"
+        return f"{b / 2**20:.2f} MiB"
     if unit == "GiB":
-        return f"{b/2**30:.2f} GiB"
+        return f"{b / 2**30:.2f} GiB"
     return f"{b:.2f} bytes"
 
 

torch/distributed/tensor/examples/convnext_example.py

@@ -243,13 +243,16 @@ def train_convnext_example():
     max_reserved = torch.cuda.max_memory_reserved()
     max_allocated = torch.cuda.max_memory_allocated()
     print(
-        f"rank {rank}, {ITER_TIME} iterations, average latency {(end - start)/ITER_TIME*1000:10.2f} ms"
+        f"rank {rank}, {ITER_TIME} iterations, "
+        f"average latency {(end - start) / ITER_TIME * 1000:10.2f} ms"
     )
     print(
-        f"rank {rank}, forward {forward_time/ITER_TIME*1000:10.2f} ms, backward {backward_time/ITER_TIME*1000:10.2f} ms"
+        f"rank {rank}, forward {forward_time / ITER_TIME * 1000:10.2f} ms, "
+        f"backward {backward_time / ITER_TIME * 1000:10.2f} ms"
     )
     print(
-        f"rank {rank}, max reserved {max_reserved/1024/1024/1024:8.2f} GiB, max allocated {max_allocated/1024/1024/1024:8.2f} GiB"
+        f"rank {rank}, max reserved {max_reserved / 1024 / 1024 / 1024:8.2f} GiB, "
+        f"max allocated {max_allocated / 1024 / 1024 / 1024:8.2f} GiB"
     )
     dist.destroy_process_group()
 

torch/distributions/utils.py

@@ -187,7 +187,7 @@ def tril_matrix_to_vec(mat: Tensor, diag: int = 0) -> Tensor:
     """
     n = mat.shape[-1]
     if not torch._C._get_tracing_state() and (diag < -n or diag >= n):
-        raise ValueError(f"diag ({diag}) provided is outside [{-n}, {n-1}].")
+        raise ValueError(f"diag ({diag}) provided is outside [{-n}, {n - 1}].")
     arange = torch.arange(n, device=mat.device)
     tril_mask = arange < arange.view(-1, 1) + (diag + 1)
     vec = mat[..., tril_mask]

torch/distributions/wishart.py

@@ -107,7 +107,7 @@ class Wishart(ExponentialFamily):
 
         if self.df.le(event_shape[-1] - 1).any():
             raise ValueError(
-                f"Value of df={df} expected to be greater than ndim - 1 = {event_shape[-1]-1}."
+                f"Value of df={df} expected to be greater than ndim - 1 = {event_shape[-1] - 1}."
             )
 
         if scale_tril is not None:

torch/export/unflatten.py

@@ -890,7 +890,7 @@ def _add_submodule(
 def _call_name(base: str, n: int) -> str:
     # Given n >= 0, generate call names to a submodule `base` of the form
     # `base`, `base@1`, `base@2`, etc.
-    return base if n == 1 else f"{base}@{n-1}"
+    return base if n == 1 else f"{base}@{n - 1}"
 
 
 def _is_call_name(call_name: str, base: str) -> bool:

torch/profiler/_memory_profiler.py

@@ -1168,8 +1168,8 @@ class MemoryProfileTimeline:
         title = "\n\n".join(
             ([title] if title else [])
             + [
-                f"Max memory allocated: {max_memory_allocated/(1024**3):.2f} GiB \n"
-                f"Max memory reserved: {max_memory_reserved/(1024**3):.2f} GiB"
+                f"Max memory allocated: {max_memory_allocated / (1024**3):.2f} GiB \n"
+                f"Max memory reserved: {max_memory_reserved / (1024**3):.2f} GiB"
             ]
         )
         axes.set_title(title)

torch/profiler/_pattern_matcher.py

@@ -84,7 +84,7 @@ class Pattern:
         )
         return (
             f"{self.name}: {len(events)} events matched. "
-            f"Total Estimated Speedup: {format_time(original_time - new_time)} ({round(original_time/new_time, 2)}X)"
+            f"Total Estimated Speedup: {format_time(original_time - new_time)} ({round(original_time / new_time, 2)}X)"
         )
 
     def match(self, event: _ProfilerEvent):
@@ -622,7 +622,7 @@ def report_all_anti_patterns(
     ]
     reported = set()
     summaries = []
-    message_list = [f"{'-'*40}TorchTidy Report{'-'*40}"]
+    message_list = [f"{'-' * 40}TorchTidy Report{'-' * 40}"]
     message_list.append("Matched Events:")
 
     for anti_pattern in anti_patterns:
@@ -657,6 +657,6 @@ def report_all_anti_patterns(
 
     message_list.append("Summary:")
     message_list += summaries
-    message_list.append(f"{'-'*40}TorchTidy Report{'-'*40}")
+    message_list.append(f"{'-' * 40}TorchTidy Report{'-' * 40}")
     if print_enable:
         print("\n".join(message_list))

torch/profiler/_utils.py

@@ -335,11 +335,11 @@ class BasicEvaluation:
 
         output += "\n".join(
             [
-                f"""{'-'*80}
+                f"""{'-' * 80}
 Event:                {event}
 Source code location: {source_code_location(event.event)}
 Percentage idle time: {self.metrics[event].fraction_idle_time * 100:.2f}%
-{'-'*80}"""
+{'-' * 80}"""
                 for event in event_list
             ]
         )

torch/testing/_internal/distributed/rpc/examples/parameter_server_test.py

@@ -112,7 +112,7 @@ def run_ps(trainers):
     torch.futures.wait_all(futs)
     stop = perf_counter()
     timed_log("Finish training")
-    timed_log(f"Time spent training: {stop-start}s")
+    timed_log(f"Time spent training: {stop - start}s")
 
 class ParameterServerTest(RpcAgentTestFixture):