[BE][Easy][3/19] enforce style for empty lines in import segments in benchmarks/ (#129754)

See https://github.com/pytorch/pytorch/pull/129751#issue-2380881501. Most changes are auto-generated by linter.

You can review these PRs via:

```bash
git diff --ignore-all-space --ignore-blank-lines HEAD~1
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/129754
Approved by: https://github.com/ezyang

benchmarks/compare-fastrnn-results.py

@@ -2,6 +2,7 @@ import argparse
 import json
 from collections import namedtuple
 
+
 Result = namedtuple("Result", ["name", "base_time", "diff_time"])
 
 

benchmarks/distributed/rpc/parameter_server/data/__init__.py

@@ -1,3 +1,4 @@
 from .DummyData import DummyData
 
+
 data_map = {"DummyData": DummyData}

benchmarks/distributed/rpc/parameter_server/models/__init__.py

@@ -1,3 +1,4 @@
 from .DummyModel import DummyModel
 
+
 model_map = {"DummyModel": DummyModel}

benchmarks/distributed/rpc/parameter_server/server/__init__.py

@@ -1,5 +1,6 @@
 from .server import AverageBatchParameterServer, AverageParameterServer
 
+
 server_map = {
     "AverageParameterServer": AverageParameterServer,
     "AverageBatchParameterServer": AverageBatchParameterServer,

benchmarks/distributed/rpc/parameter_server/trainer/__init__.py

@@ -6,6 +6,7 @@ from .iteration_steps import basic_iteration_step
 from .preprocess_data import preprocess_dummy_data
 from .trainer import DdpTrainer
 
+
 criterion_map = {"cel": cel}
 
 ddp_hook_map = {

benchmarks/distributed/rpc/parameter_server/utils.py

@@ -1,5 +1,6 @@
 import torch
 
+
 RPC_SPARSE = "rpc_sparse"
 RPC_DENSE = "rpc_dense"
 

benchmarks/distributed/rpc/rl/coordinator.py

@@ -1,13 +1,13 @@
 import time
 
 import numpy as np
-
 from agent import AgentBase
 from observer import ObserverBase
 
 import torch
 import torch.distributed.rpc as rpc
 
+
 COORDINATOR_NAME = "coordinator"
 AGENT_NAME = "agent"
 OBSERVER_NAME = "observer{}"

benchmarks/distributed/rpc/rl/launcher.py

@@ -1,5 +1,4 @@
 import argparse
-
 import json
 import os
 import time
@@ -9,6 +8,7 @@ from coordinator import CoordinatorBase
 import torch.distributed.rpc as rpc
 import torch.multiprocessing as mp
 
+
 COORDINATOR_NAME = "coordinator"
 AGENT_NAME = "agent"
 OBSERVER_NAME = "observer{}"

benchmarks/dynamo/benchmarks.py

@@ -1,8 +1,8 @@
 #!/usr/bin/env python3
+
 import argparse
 import os
 import sys
-
 from typing import Set
 
 

benchmarks/dynamo/ci_expected_accuracy/update_expected.py

@@ -30,6 +30,7 @@ from zipfile import ZipFile
 import pandas as pd
 import requests
 
+
 # Note: the public query url targets this rockset lambda:
 # https://console.rockset.com/lambdas/details/commons.artifacts
 ARTIFACTS_QUERY_URL = "https://api.usw2a1.rockset.com/v1/public/shared_lambdas/4ca0033e-0117-41f5-b043-59cde19eff35"

benchmarks/dynamo/combine_csv.py

@@ -6,6 +6,7 @@ import csv
 import sys
 from collections import defaultdict
 
+
 assert len(sys.argv) == 3
 
 RESULTS = defaultdict(dict)

benchmarks/dynamo/common.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+
 from __future__ import annotations
 
 import abc

benchmarks/dynamo/dist_util.py

@@ -15,6 +15,7 @@ from torch.distributed.algorithms._checkpoint.checkpoint_wrapper import (
 from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
 from torch.distributed.fsdp.wrap import ModuleWrapPolicy
 
+
 try:
     from .torchbench import setup_torchbench_cwd
 except ImportError:

benchmarks/dynamo/distributed.py

@@ -10,6 +10,7 @@ from torch._dynamo.testing import reduce_to_scalar_loss
 from torch.nn.parallel import DistributedDataParallel as DDP
 from torch.profiler import profile, ProfilerActivity, record_function
 
+
 try:
     from .common import timed
     from .dist_util import apply_fsdp, cleanup, get_model, model_iter_fn, setup

benchmarks/dynamo/huggingface.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+
 import importlib
 import logging
 import os
@@ -7,16 +8,17 @@ import subprocess
 import sys
 import warnings
 
+
 try:
     from .common import BenchmarkRunner, download_retry_decorator, main, reset_rng_state
 except ImportError:
     from common import BenchmarkRunner, download_retry_decorator, main, reset_rng_state
 
 import torch
-
 from torch._dynamo.testing import collect_results
 from torch._dynamo.utils import clone_inputs
 
+
 log = logging.getLogger(__name__)
 
 # Enable FX graph caching

benchmarks/dynamo/microbenchmarks/bench_mm_fusion.py

@@ -1,12 +1,13 @@
 # flake8: noqa
+
 import triton
 from prettytable import PrettyTable
 
 import torch
-
 import torch._dynamo
 import torch._inductor.config
 
+
 # torch._inductor.config.debug = True
 torch._inductor.config.triton.dense_indexing = True
 torch.manual_seed(0)

benchmarks/dynamo/microbenchmarks/fx_microbenchmarks.py

@@ -2,6 +2,7 @@ import timeit
 
 import torch.fx
 
+
 N = 100000
 K = 1000
 

benchmarks/dynamo/microbenchmarks/inductor_bmm.py

@@ -1,7 +1,6 @@
 from benchmark_helper import time_with_torch_timer
 
 import torch
-
 import torch._dynamo
 import torch._dynamo.config
 import torch._inductor.config as config

benchmarks/dynamo/microbenchmarks/inductor_mm.py

@@ -2,11 +2,11 @@ import triton
 from benchmark_helper import time_with_torch_timer
 
 import torch
-
 import torch._dynamo
 import torch._dynamo.config
 import torch._inductor.config as config
 
+
 # The flag below controls whether to allow TF32 on matmul. This flag defaults to True.
 torch.backends.cuda.matmul.allow_tf32 = True
 # The flag below controls whether to allow TF32 on cuDNN. This flag defaults to True.

benchmarks/dynamo/microbenchmarks/matmul_relu.py

@@ -1,10 +1,10 @@
 from benchmark_helper import time_with_torch_timer
 
 import torch
-
 import torch._dynamo
 import torch._inductor.config as inductor_config
 
+
 inductor_config.triton.mm = "triton"
 
 

benchmarks/dynamo/microbenchmarks/microbench.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+
 import argparse
 import inspect
 import sys
@@ -7,13 +8,13 @@ import numpy as np
 import tabulate
 
 import torch
-
 import torch._inductor
 from torch._dynamo.backends.cudagraphs import cudagraphs_inner
 from torch._dynamo.testing import same
 from torch._inductor.compile_fx import compile_fx
 from torch._inductor.utils import timed
 
+
 aten = torch.ops.aten
 
 try:

benchmarks/dynamo/microbenchmarks/operator_inp_utils.py

@@ -12,6 +12,7 @@ from torch.utils import _pytree as pytree
 from torch.utils._python_dispatch import TorchDispatchMode
 from torch.utils._pytree import tree_map
 
+
 log = logging.getLogger(__name__)
 
 OP_INP_DIRECTORY = os.path.join(os.path.dirname(__file__), "operator_inp_logs")

benchmarks/dynamo/microbenchmarks/operatorbench.py

@@ -1,10 +1,10 @@
 #!/usr/bin/env python3
+
 import click
 import numpy as np
 from operator_inp_utils import OperatorInputsLoader
 
 import torch
-
 from torch._dynamo.backends.cudagraphs import cudagraphs_inner
 from torch._dynamo.testing import same
 from torch._inductor.compile_fx import compile_fx
@@ -13,6 +13,7 @@ from torch._inductor.lowering import lowerings
 from torch._inductor.utils import gen_gm_and_inputs
 from torch.utils._pytree import tree_map_only
 
+
 aten = torch.ops.aten
 
 

benchmarks/dynamo/parse_logs.py

@@ -3,6 +3,7 @@ import os
 import re
 import sys
 
+
 # This script takes the logs produced by the benchmark scripts (e.g.,
 # torchbench.py) and parses it into a CSV file that summarizes what
 # is failing and why.  It is kept separate from the benchmark script

benchmarks/dynamo/runner.py

@@ -23,7 +23,6 @@ If you want to test float16
 
 """
 
-
 import argparse
 import dataclasses
 import functools
@@ -44,7 +43,6 @@ from os.path import abspath, exists
 from random import randint
 
 import matplotlib.pyplot as plt
-
 import numpy as np
 import pandas as pd
 from matplotlib import rcParams
@@ -52,9 +50,9 @@ from scipy.stats import gmean
 from tabulate import tabulate
 
 import torch
-
 import torch._dynamo
 
+
 rcParams.update({"figure.autolayout": True})
 plt.rc("axes", axisbelow=True)
 

benchmarks/dynamo/test.py

@@ -2,9 +2,9 @@ import os
 import unittest
 
 from .common import parse_args, run
-
 from .torchbench import setup_torchbench_cwd, TorchBenchmarkRunner
 
+
 try:
     # fbcode only
     from aiplatform.utils.sanitizer_status import is_asan_or_tsan

benchmarks/dynamo/timm_models.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+
 import importlib
 import logging
 import os
@@ -7,16 +8,17 @@ import subprocess
 import sys
 import warnings
 
+
 try:
     from .common import BenchmarkRunner, download_retry_decorator, main
 except ImportError:
     from common import BenchmarkRunner, download_retry_decorator, main
 
 import torch
-
 from torch._dynamo.testing import collect_results, reduce_to_scalar_loss
 from torch._dynamo.utils import clone_inputs
 
+
 # Enable FX graph caching
 if "TORCHINDUCTOR_FX_GRAPH_CACHE" not in os.environ:
     torch._inductor.config.fx_graph_cache = True

benchmarks/dynamo/torchbench.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+
 import functools
 import gc
 import importlib
@@ -14,6 +15,7 @@ import yaml
 
 import torch
 
+
 try:
     from .common import BenchmarkRunner, main
 except ImportError:
@@ -22,6 +24,7 @@ except ImportError:
 from torch._dynamo.testing import collect_results, reduce_to_scalar_loss
 from torch._dynamo.utils import clone_inputs
 
+
 # We are primarily interested in tf32 datatype
 torch.backends.cuda.matmul.allow_tf32 = True
 

benchmarks/dynamo/training_loss.py

@@ -9,10 +9,10 @@ from datasets import load_dataset, load_metric
 from transformers import AutoModelForSequenceClassification, AutoTokenizer
 
 import torch
-
 import torch._dynamo
 from torch.utils.data import DataLoader
 
+
 torch.backends.cuda.matmul.allow_tf32 = True
 
 # You will download around 84G dataset if you run this end to end training/evaluation example.

benchmarks/fastrnns/__init__.py

@@ -1,6 +1,7 @@
 from .cells import *  # noqa: F403
 from .factory import *  # noqa: F403
 
+
 # (output, next_state) = cell(input, state)
 seqLength = 100
 numLayers = 2

benchmarks/fastrnns/conftest.py

@@ -1,5 +1,6 @@
 import pytest  # noqa: F401
 
+
 default_rnns = [
     "cudnn",
     "aten",

benchmarks/fastrnns/custom_lstms.py

@@ -9,6 +9,7 @@ import torch.nn as nn
 from torch import Tensor
 from torch.nn import Parameter
 
+
 """
 Some helper classes for writing custom TorchScript LSTMs.
 

benchmarks/fastrnns/scratch.py

@@ -45,6 +45,7 @@ recurrent_scaleshift.graph_for(x, scale, shift)
 
 import torch
 
+
 x = torch.tensor([])
 x.requires_grad = True
 x.mean().backward()  # no error triggered

benchmarks/framework_overhead_benchmark/framework_overhead_benchmark.py

@@ -1,10 +1,11 @@
 import argparse
 
 from pt_wrapper_module import WrapperModule
-
 from SimpleAddModule import add_tensors_loop, SimpleAddModule
+
 from utils import benchmark_module, BenchmarkConfig, ModuleConfig, ms_to_us
 
+
 """ Framework overhead benchmark script.
 Benchmark framework overhead.
 Currently supported ops: add.

benchmarks/framework_overhead_benchmark/utils.py

@@ -3,6 +3,7 @@ from collections import namedtuple
 
 from torch.utils import ThroughputBenchmark
 
+
 NUM_LOOP_ITERS = 1000
 BenchmarkConfig = namedtuple("BenchmarkConfig", "num_warmup_iters num_iters")
 ModuleConfig = namedtuple("ModuleConfig", "pt_fn c2_op num_params graph_mode")

benchmarks/functional_autograd_benchmark/audio_text_models.py

@@ -1,5 +1,4 @@
 import torchaudio_models as models
-
 from utils import check_for_functorch, extract_weights, GetterReturnType, load_weights
 
 import torch

benchmarks/functional_autograd_benchmark/functional_autograd_benchmark.py

@@ -6,6 +6,7 @@ from typing import Any, Callable, List, NamedTuple
 import torch
 from torch.autograd import functional
 
+
 try:
     import functorch as ft
 

benchmarks/functional_autograd_benchmark/torchaudio_models.py

@@ -9,6 +9,7 @@ import torch
 import torch.nn.functional as F
 from torch import nn, Tensor
 
+
 __all__ = ["Wav2Letter"]
 
 

benchmarks/functional_autograd_benchmark/torchvision_models.py

@@ -4,10 +4,10 @@ from collections import OrderedDict
 
 import torch
 from torch import nn
-
 from torch.jit.annotations import Dict
 from torch.nn import functional as F
 
+
 try:
     from scipy.optimize import linear_sum_assignment
 

benchmarks/functional_autograd_benchmark/utils.py

@@ -2,9 +2,9 @@ from collections import defaultdict
 from typing import Callable, Dict, List, Tuple, Union
 
 import torch
-
 from torch import nn, Tensor
 
+
 # Type helpers
 InputsType = Union[Tensor, Tuple[Tensor, ...]]
 # A Getter takes in a device and returns a callable and the inputs to that callable

benchmarks/functional_autograd_benchmark/vision_models.py

@@ -1,12 +1,12 @@
 from typing import cast
 
 import torchvision_models as models
-
 from utils import check_for_functorch, extract_weights, GetterReturnType, load_weights
 
 import torch
 from torch import Tensor
 
+
 has_functorch = check_for_functorch()
 
 

benchmarks/fuser/plot_speedups.py

@@ -1,5 +1,6 @@
 import pandas
 
+
 df = pandas.read_csv("perf.csv")
 
 ops = pandas.unique(df["operator"])
@@ -11,6 +12,7 @@ pivot_speedups = (pivot_op_shape.T / pivot_op_shape["eager"]).T
 
 import matplotlib.pyplot as plt
 
+
 plt.rcParams["figure.figsize"] = (20, 100)
 fig, axs = plt.subplots(nops)
 plt.subplots_adjust(hspace=0.5)

benchmarks/fuser/run_benchmarks.py

@@ -7,6 +7,7 @@ import click
 
 import torch
 
+
 torch.set_num_threads(1)
 torch._C._debug_set_fusion_group_inlining(False)
 

benchmarks/gpt_fast/benchmark.py

@@ -10,6 +10,7 @@ import torch
 import torch.nn as nn
 from torch.utils.flop_counter import FlopCounterMode
 
+
 WARMUP_ITER = 5
 
 A100_40G_BF16_TFLOPS = 312

benchmarks/gpt_fast/generate.py

@@ -14,6 +14,7 @@ from quantize import WeightOnlyInt8QuantHandler as LLaMAWeightOnlyInt8QuantHandl
 import torch
 import torch._inductor.config
 
+
 torch._inductor.config.coordinate_descent_tuning = True
 torch._inductor.config.triton.unique_kernel_names = True
 torch._inductor.config.fx_graph_cache = True  # Experimental feature to reduce compilation times, will be on by default in future

benchmarks/gpt_fast/mixtral_moe_quantize.py

@@ -5,6 +5,7 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
+
 ##### Quantization Primitives ######
 
 

benchmarks/inference/process_metrics.py

@@ -10,6 +10,7 @@ import os
 
 import pandas as pd
 
+
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Parse output files")
     parser.add_argument("--csv", type=str, help="Path to csv file")

benchmarks/inference/server.py

@@ -1,5 +1,4 @@
 import argparse
-
 import asyncio
 import os.path
 import subprocess

benchmarks/instruction_counts/core/api.py

@@ -8,6 +8,7 @@ from typing import Dict, List, Optional, Set, Tuple, TYPE_CHECKING, Union
 
 from worker.main import WorkerTimerArgs
 
+
 if TYPE_CHECKING:
     # Benchmark utils are only partially strict compliant, so MyPy won't follow
     # imports using the public namespace. (Due to an exclusion rule in

benchmarks/instruction_counts/core/expand.py

@@ -13,6 +13,7 @@ from typing import List, Optional, Tuple, TYPE_CHECKING
 
 import torch
 
+
 if TYPE_CHECKING:
     # See the note in api.py for why this is necessary.
     from torch.utils.benchmark.utils.timer import Language

benchmarks/instruction_counts/execution/work.py

@@ -20,6 +20,7 @@ from worker.main import (
     WorkerUnpickler,
 )
 
+
 if TYPE_CHECKING:
     PopenType = subprocess.Popen[bytes]
 else:

benchmarks/operator_benchmark/benchmark_all_other_test.py

@@ -1,4 +1,4 @@
-from pt import (  # noqa: F401  # noqa: F401
+from pt import (  # noqa: F401
     add_test,
     ao_sparsifier_test,
     as_strided_test,
@@ -31,5 +31,6 @@ from pt import (  # noqa: F401  # noqa: F401
 
 import operator_benchmark as op_bench
 
+
 if __name__ == "__main__":
     op_bench.benchmark_runner.main()

benchmarks/operator_benchmark/benchmark_all_test.py

@@ -4,5 +4,6 @@ from pt import unary_test  # noqa: F401
 
 import operator_benchmark as op_bench
 
+
 if __name__ == "__main__":
     op_bench.benchmark_runner.main()

benchmarks/operator_benchmark/benchmark_core.py

@@ -6,6 +6,7 @@ import timeit
 from collections import namedtuple
 
 import benchmark_utils
+
 import numpy as np
 
 import torch

benchmarks/operator_benchmark/benchmark_runner.py

@@ -1,10 +1,12 @@
 import argparse
 
 import benchmark_core
+
 import benchmark_utils
 
 import torch
 
+
 """Performance microbenchmarks's main binary.
 
 This is the main function for running performance microbenchmark tests.

benchmarks/operator_benchmark/common/repeat_benchmark.py

@@ -4,6 +4,7 @@ import numpy as np
 
 import torch
 
+
 """Microbenchmarks for Tensor repeat operator. Supports PyTorch."""
 
 input_shapes = (

benchmarks/operator_benchmark/common/tests/jit_forward_test.py

@@ -2,6 +2,7 @@ import operator_benchmark as op_bench
 
 import torch
 
+
 intraop_bench_configs = op_bench.config_list(
     attrs=[
         [8, 16],

benchmarks/operator_benchmark/common/tests/pt_configs_list_test.py

@@ -2,6 +2,7 @@ import operator_benchmark as op_bench
 
 import torch
 
+
 """Microbenchmarks for element-wise Add operator. Supports both Caffe2/PyTorch."""
 
 add_short_configs = op_bench.config_list(

benchmarks/operator_benchmark/operator_benchmark.py

@@ -2,4 +2,5 @@
 import benchmark_runner  # noqa: F401
 from benchmark_pytorch import TorchBenchmarkBase  # noqa: F401
 from benchmark_test_generator import *  # noqa: F401,F403
+
 from benchmark_utils import *  # noqa: F401,F403

benchmarks/operator_benchmark/pt/add_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for add_ operator. Supports both Caffe2/PyTorch."""
 
 # Configs for PT add operator

benchmarks/operator_benchmark/pt/ao_sparsifier_test.py

@@ -1,7 +1,7 @@
 import operator_benchmark as op_bench
+
 import torch
 from torch import nn
-
 from torch.ao import pruning
 
 

benchmarks/operator_benchmark/pt/batchnorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn.functional as F
 

benchmarks/operator_benchmark/pt/binary_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/bmm_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for add_ operator. Supports both Caffe2/PyTorch."""
 
 

benchmarks/operator_benchmark/pt/channel_shuffle_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/chunk_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/clip_ranges_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/configs.py

@@ -1,5 +1,6 @@
 import operator_benchmark as op_bench
 
+
 """
 Configs shared by multiple benchmarks
 """

benchmarks/operator_benchmark/pt/conv_test.py

@@ -1,9 +1,11 @@
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn as nn
 
+
 """
 Microbenchmarks for Conv1d and ConvTranspose1d operators.
 """

benchmarks/operator_benchmark/pt/diag_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/embeddingbag_test.py

@@ -2,8 +2,10 @@ import numpy
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Embedding and EmbeddingBag Operator Benchmark"""
 
 

benchmarks/operator_benchmark/pt/fill_test.py

@@ -1,8 +1,9 @@
 import operator_benchmark as op_bench
-import torch
 
+import torch
 from torch.testing._internal.common_device_type import get_all_device_types
 
+
 """Microbenchmark for Fill_ operator."""
 
 fill_short_configs = op_bench.config_list(

benchmarks/operator_benchmark/pt/gather_test.py

@@ -1,6 +1,7 @@
 import numpy
 
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/gelu_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/groupnorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn.functional as F
 

benchmarks/operator_benchmark/pt/hardsigmoid_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn as nn
 

benchmarks/operator_benchmark/pt/hardswish_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn as nn
 

benchmarks/operator_benchmark/pt/index_select_test.py

@@ -1,6 +1,7 @@
 import numpy
 
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/instancenorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn.functional as F
 

benchmarks/operator_benchmark/pt/interpolate_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for interpolate operator."""
 
 

benchmarks/operator_benchmark/pt/layernorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn.functional as F
 

benchmarks/operator_benchmark/pt/linear_prepack_fp16_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for linear_prepack_fp16_ operator. Supports both Caffe2/PyTorch."""
 
 # Configs for PT linear_prepack_fp16 operator

benchmarks/operator_benchmark/pt/linear_test.py

@@ -1,6 +1,7 @@
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn as nn
 

benchmarks/operator_benchmark/pt/linear_unpack_fp16_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for linear_unpack_fp16_ operator. Supports both Caffe2/PyTorch."""
 
 # Configs for PT linear_unpack_fp16 operator

benchmarks/operator_benchmark/pt/matmul_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for MatMul operator"""
 
 # Configs for PT Matmul operator

benchmarks/operator_benchmark/pt/matrix_mult_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """
 Microbenchmarks for batch matrix mult with einsum and torch.bmm.
 """

benchmarks/operator_benchmark/pt/pool_test.py

@@ -1,7 +1,9 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.nn as nn
 
+
 """
 Microbenchmarks for MaxPool1d and AvgPool1d operators.
 """

benchmarks/operator_benchmark/pt/qactivation_test.py

@@ -1,7 +1,9 @@
 import operator_benchmark as op_bench
+
 import torch
 import torch.ao.nn.quantized.functional as qF
 
+
 r"""Microbenchmarks for the quantized activations."""
 
 qactivation_long_configs = op_bench.cross_product_configs(

benchmarks/operator_benchmark/pt/qarithmetic_test.py

@@ -1,7 +1,9 @@
 import operator_benchmark as op_bench
+
 import torch
 from torch._ops import ops
 
+
 qarithmetic_binary_configs = op_bench.cross_product_configs(
     N=(2, 8, 64, 512),
     dtype=(torch.quint8, torch.qint8, torch.qint32),

benchmarks/operator_benchmark/pt/qatembedding_ops_test.py

@@ -2,10 +2,12 @@ import numpy
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 import torch.ao.nn.qat as nnqat
 from torch.ao.quantization import default_embedding_qat_qconfig
 
+
 """
 Microbenchmarks for QAT Embedding + EmbeddingBag operators.
 """

benchmarks/operator_benchmark/pt/qbatchnorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/qcomparators_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 qcomparators_configs = op_bench.cross_product_configs(
     N=(8, 64),
     dtype=(torch.quint8, torch.qint8, torch.qint32),

benchmarks/operator_benchmark/pt/qconv_test.py

@@ -1,9 +1,11 @@
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 import torch.ao.nn.quantized as nnq
 
+
 """
 Microbenchmarks for qConv operators.
 """

benchmarks/operator_benchmark/pt/qembedding_bag_lookups_test.py

@@ -3,10 +3,11 @@ from typing import Optional
 import numpy as np
 
 import operator_benchmark as op_bench
-import torch
 
+import torch
 from torch.testing._internal.common_quantization import lengths_to_offsets
 
+
 torch.ops.load_library("//caffe2/torch/fb/sparsenn:sparsenn_operators")
 
 

benchmarks/operator_benchmark/pt/qembedding_pack_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 embeddingbag_conversion_short_configs = op_bench.cross_product_configs(
     num_embeddings=(80,), embedding_dim=(128, 256, 512), tags=("short",)
 )

benchmarks/operator_benchmark/pt/qembeddingbag_test.py

@@ -2,9 +2,11 @@ import numpy
 from pt import configs
 
 import operator_benchmark as op_bench
+
 import torch
 import torch.ao.nn.quantized as nnq
 
+
 """
 Microbenchmarks for qEmbeddingBag operators.
 """

benchmarks/operator_benchmark/pt/qgroupnorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/qinstancenorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch
 
 

benchmarks/operator_benchmark/pt/qinterpolate_test.py

@@ -1,6 +1,8 @@
 import operator_benchmark as op_bench
+
 import torch
 
+
 """Microbenchmarks for the quantized interpolate op.
 
 Note: We are not benchmarking `upsample` as it is being deprecated, and calls

benchmarks/operator_benchmark/pt/qlayernorm_test.py

@@ -1,4 +1,5 @@
 import operator_benchmark as op_bench
+
 import torch