Revert "[5/N][Easy] fix typo for usort config in pyproject.toml (kown -> known): sort torch (#127126)"

This reverts commit 7763c83af67eebfdd5185dbe6ce15ece2b992a0f.

Reverted https://github.com/pytorch/pytorch/pull/127126 on behalf of https://github.com/XuehaiPan due to Broken CI ([comment](https://github.com/pytorch/pytorch/pull/127126#issuecomment-2133044286))

android/test_app/make_assets.py

@@ -1,6 +1,5 @@
-from torchvision import models
-
 import torch
+from torchvision import models
 
 print(torch.version.__version__)
 

android/test_app/make_assets_custom.py

@@ -4,11 +4,10 @@ MobileNetV2 TorchScript model, and dumps root ops used by the model for custom
 build script to create a tailored build which only contains these used ops.
 """
 
+import torch
 import yaml
 from torchvision import models
 
-import torch
-
 # Download and trace the model.
 model = models.mobilenet_v2(weights=models.MobileNet_V2_Weights.IMAGENET1K_V1)
 model.eval()

benchmarks/distributed/ddp/benchmark.py

@@ -18,12 +18,11 @@ import sys
 import time
 
 import numpy as np
-import torchvision
-
 import torch
 import torch.distributed as dist
 import torch.nn as nn
 import torch.optim as optim
+import torchvision
 
 
 def allgather_object(obj):

benchmarks/distributed/pipeline/pipe.py

@@ -3,10 +3,10 @@ import math
 import os
 import time
 
-from benchmark_dataset import BenchmarkLMDataset, collate_sentences_lm
-
 import torch
 import torch.nn as nn
+
+from benchmark_dataset import BenchmarkLMDataset, collate_sentences_lm
 from torch.distributed import rpc
 
 from torch.distributed.pipeline.sync import Pipe

benchmarks/distributed/rpc/parameter_server/launcher.py

@@ -3,10 +3,18 @@ import json
 import os
 from pathlib import Path
 
+import torch
+import torch.distributed as c10d
+import torch.distributed.rpc as rpc
+import torch.multiprocessing as mp
+
 from data import data_map
 from metrics.ProcessedMetricsPrinter import ProcessedMetricsPrinter
 from models import model_map
 from server import server_map
+from torch.distributed.rpc import TensorPipeRpcBackendOptions
+from torch.futures import wait_all
+from torch.utils.data import DataLoader
 from trainer import (
     criterion_map,
     ddp_hook_map,
@@ -17,14 +25,6 @@ from trainer import (
     trainer_map,
 )
 
-import torch
-import torch.distributed as c10d
-import torch.distributed.rpc as rpc
-import torch.multiprocessing as mp
-from torch.distributed.rpc import TensorPipeRpcBackendOptions
-from torch.futures import wait_all
-from torch.utils.data import DataLoader
-
 
 def get_name(rank, args):
     r"""

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

@@ -3,12 +3,12 @@ import threading
 import time
 from abc import ABC, abstractmethod
 
-from metrics.MetricsLogger import MetricsLogger
-from utils import sparse_rpc_format_to_tensor, sparse_tensor_to_rpc_format
-
 import torch
 import torch.distributed.rpc as rpc
 
+from metrics.MetricsLogger import MetricsLogger
+from utils import sparse_rpc_format_to_tensor, sparse_tensor_to_rpc_format
+
 
 class ParameterServerBase(ABC):
     PARAMETER_SERVER_BATCH_METRIC = "parameter_server_batch_metric"

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

@@ -1,7 +1,6 @@
-from utils import process_bucket_with_remote_server
-
 import torch
 import torch.distributed as c10d
+from utils import process_bucket_with_remote_server
 
 
 def allreduce_hook(state, bucket):

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

@@ -2,10 +2,10 @@ import functools
 import time
 from abc import ABC, abstractmethod
 
-from metrics.MetricsLogger import MetricsLogger
-
 import torch
 
+from metrics.MetricsLogger import MetricsLogger
+
 
 class TrainerBase(ABC):
     BATCH_LEVEL_METRIC = "batch_level_metric"

benchmarks/distributed/rpc/rl/coordinator.py

@@ -2,12 +2,12 @@ import time
 
 import numpy as np
 
-from agent import AgentBase
-from observer import ObserverBase
-
 import torch
 import torch.distributed.rpc as rpc
 
+from agent import AgentBase
+from observer import ObserverBase
+
 COORDINATOR_NAME = "coordinator"
 AGENT_NAME = "agent"
 OBSERVER_NAME = "observer{}"

benchmarks/distributed/rpc/rl/launcher.py

@@ -4,11 +4,11 @@ import json
 import os
 import time
 
-from coordinator import CoordinatorBase
-
 import torch.distributed.rpc as rpc
 import torch.multiprocessing as mp
 
+from coordinator import CoordinatorBase
+
 COORDINATOR_NAME = "coordinator"
 AGENT_NAME = "agent"
 OBSERVER_NAME = "observer{}"

benchmarks/distributed/rpc/rl/observer.py

@@ -1,10 +1,10 @@
 import random
 import time
 
-from agent import AgentBase
-
 import torch
 import torch.distributed.rpc as rpc
+
+from agent import AgentBase
 from torch.distributed.rpc import rpc_sync
 
 

benchmarks/dynamo/common.py

@@ -36,21 +36,20 @@ from typing import (
     Type,
     TYPE_CHECKING,
 )
+
 from typing_extensions import Self
 from unittest.mock import MagicMock
 
 import numpy as np
 import pandas as pd
 import psutil
-from scipy.stats import gmean, ttest_ind
-from tqdm.auto import tqdm, trange
-
 import torch
 import torch._dynamo
 import torch._dynamo.utils
 import torch._export
 import torch.distributed
 import torch.multiprocessing as mp
+from scipy.stats import gmean, ttest_ind
 from torch._C import _has_cuda as HAS_CUDA, _has_xpu as HAS_XPU
 from torch._dynamo.profiler import fx_insert_profiling, Profiler
 from torch._dynamo.testing import (
@@ -60,6 +59,8 @@ from torch._dynamo.testing import (
     same,
 )
 
+from tqdm.auto import tqdm, trange
+
 try:
     from torch._dynamo.utils import (
         clone_inputs,
@@ -73,14 +74,15 @@ except ImportError:
         graph_break_reasons,
         maybe_enable_compiled_autograd,
     )
-
 import torch._functorch.config
 from torch._functorch.aot_autograd import set_model_name
 from torch._inductor import config as inductor_config, metrics
 from torch._subclasses.fake_tensor import FakeTensorMode
+
 from torch.utils import _pytree as pytree
 from torch.utils._pytree import tree_map, tree_map_only
 
+
 try:
     import torch_xla
     import torch_xla.core.xla_model as xm
@@ -2341,16 +2343,16 @@ class BenchmarkRunner:
 
     def get_fsdp_auto_wrap_policy(self, model_name: str):
         from diffusers.models.transformer_2d import Transformer2DModel
-        from torchbenchmark.models.nanogpt.model import Block
-        from transformers.models.llama.modeling_llama import LlamaDecoderLayer
-
-        from transformers.models.t5.modeling_t5 import T5Block
-        from transformers.models.whisper.modeling_whisper import WhisperEncoderLayer
 
         from torch.distributed.fsdp.wrap import (
             ModuleWrapPolicy,
             size_based_auto_wrap_policy,
         )
+        from torchbenchmark.models.nanogpt.model import Block
+        from transformers.models.llama.modeling_llama import LlamaDecoderLayer
+
+        from transformers.models.t5.modeling_t5 import T5Block
+        from transformers.models.whisper.modeling_whisper import WhisperEncoderLayer
 
         # handcrafted wrap policy
         MODEL_FSDP_WRAP = {

benchmarks/dynamo/huggingface.py

@@ -7,9 +7,8 @@ import subprocess
 import sys
 import warnings
 
-from common import BenchmarkRunner, download_retry_decorator, main, reset_rng_state
-
 import torch
+from common import BenchmarkRunner, download_retry_decorator, main, reset_rng_state
 
 from torch._dynamo.testing import collect_results
 from torch._dynamo.utils import clone_inputs

benchmarks/dynamo/microbenchmarks/bench_mm_fusion.py

@@ -1,11 +1,10 @@
 # flake8: noqa
-import triton
-from prettytable import PrettyTable
-
 import torch
 
 import torch._dynamo
 import torch._inductor.config
+import triton
+from prettytable import PrettyTable
 
 # torch._inductor.config.debug = True
 torch._inductor.config.triton.dense_indexing = True

benchmarks/dynamo/microbenchmarks/inductor_bmm.py

@@ -1,10 +1,9 @@
-from benchmark_helper import time_with_torch_timer
-
 import torch
 
 import torch._dynamo
 import torch._dynamo.config
 import torch._inductor.config as config
+from benchmark_helper import time_with_torch_timer
 
 
 @torch._dynamo.optimize("inductor", nopython=True)

benchmarks/dynamo/microbenchmarks/inductor_cpu_atomic.py

@@ -1,9 +1,8 @@
 import itertools
 
-from benchmark_helper import time_with_torch_timer
-
 import torch
 import torch._dynamo
+from benchmark_helper import time_with_torch_timer
 
 
 @torch._dynamo.optimize("inductor", nopython=True)

benchmarks/dynamo/microbenchmarks/inductor_mm.py

@@ -1,11 +1,10 @@
-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
+import triton
+from benchmark_helper import time_with_torch_timer
 
 # The flag below controls whether to allow TF32 on matmul. This flag defaults to True.
 torch.backends.cuda.matmul.allow_tf32 = True

benchmarks/dynamo/microbenchmarks/matmul_relu.py

@@ -1,9 +1,8 @@
-from benchmark_helper import time_with_torch_timer
-
 import torch
 
 import torch._dynamo
 import torch._inductor.config as inductor_config
+from benchmark_helper import time_with_torch_timer
 
 inductor_config.triton.mm = "triton"
 

benchmarks/dynamo/microbenchmarks/microbench.py

@@ -5,7 +5,6 @@ import sys
 
 import numpy as np
 import tabulate
-
 import torch
 
 import torch._inductor

benchmarks/dynamo/microbenchmarks/operatorbench.py

@@ -1,9 +1,8 @@
 #!/usr/bin/env python3
 import click
 import numpy as np
-from operator_inp_utils import OperatorInputsLoader
-
 import torch
+from operator_inp_utils import OperatorInputsLoader
 
 from torch._dynamo.backends.cudagraphs import cudagraphs_inner
 from torch._dynamo.testing import same

benchmarks/dynamo/runner.py

@@ -47,13 +47,12 @@ import matplotlib.pyplot as plt
 
 import numpy as np
 import pandas as pd
-from matplotlib import rcParams
-from scipy.stats import gmean
-from tabulate import tabulate
-
 import torch
 
 import torch._dynamo
+from matplotlib import rcParams
+from scipy.stats import gmean
+from tabulate import tabulate
 
 rcParams.update({"figure.autolayout": True})
 plt.rc("axes", axisbelow=True)

benchmarks/dynamo/timm_models.py

@@ -7,9 +7,8 @@ import subprocess
 import sys
 import warnings
 
-from common import BenchmarkRunner, download_retry_decorator, main
-
 import torch
+from common import BenchmarkRunner, download_retry_decorator, main
 
 from torch._dynamo.testing import collect_results, reduce_to_scalar_loss
 from torch._dynamo.utils import clone_inputs

benchmarks/dynamo/torchbench.py

@@ -10,9 +10,8 @@ import warnings
 from collections import namedtuple
 from os.path import abspath, exists
 
-import yaml
-
 import torch
+import yaml
 
 try:
     from .common import BenchmarkRunner, main

benchmarks/dynamo/training_loss.py

@@ -5,13 +5,12 @@ import sys
 import time
 from datetime import timedelta
 
-from datasets import load_dataset, load_metric
-from transformers import AutoModelForSequenceClassification, AutoTokenizer
-
 import torch
 
 import torch._dynamo
+from datasets import load_dataset, load_metric
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
 
 torch.backends.cuda.matmul.allow_tf32 = True
 

benchmarks/fastrnns/runner.py

@@ -1,9 +1,8 @@
 from collections import namedtuple
 from functools import partial
 
-import torchvision.models as cnn
-
 import torch
+import torchvision.models as cnn
 
 from .factory import (
     dropoutlstm_creator,

benchmarks/fastrnns/test_bench.py

@@ -1,5 +1,4 @@
 import pytest
-
 import torch
 
 from .fuser import set_fuser

benchmarks/framework_overhead_benchmark/SimpleAddModule.py

@@ -1,6 +1,5 @@
-from utils import NUM_LOOP_ITERS
-
 import torch
+from utils import NUM_LOOP_ITERS
 
 
 def add_tensors_loop(x, y):

benchmarks/functional_autograd_benchmark/audio_text_models.py

@@ -1,10 +1,10 @@
+import torch
+
 import torchaudio_models as models
+from torch import nn, Tensor
 
 from utils import check_for_functorch, extract_weights, GetterReturnType, load_weights
 
-import torch
-from torch import nn, Tensor
-
 
 has_functorch = check_for_functorch()
 

benchmarks/functional_autograd_benchmark/ppl_models.py

@@ -1,9 +1,9 @@
-from utils import GetterReturnType
-
 import torch
 import torch.distributions as dist
 from torch import Tensor
 
+from utils import GetterReturnType
+
 
 def get_simple_regression(device: torch.device) -> GetterReturnType:
     N = 10

benchmarks/functional_autograd_benchmark/vision_models.py

@@ -1,12 +1,11 @@
 from typing import cast
 
+import torch
 import torchvision_models as models
+from torch import Tensor
 
 from utils import check_for_functorch, extract_weights, GetterReturnType, load_weights
 
-import torch
-from torch import Tensor
-
 has_functorch = check_for_functorch()
 
 

benchmarks/fuser/run_benchmarks.py

@@ -4,7 +4,6 @@ import sys
 import time
 
 import click
-
 import torch
 
 torch.set_num_threads(1)

benchmarks/gpt_fast/benchmark.py

@@ -6,6 +6,8 @@ import os
 import time
 from typing import Optional, Tuple
 
+import torch
+import torch._inductor.config
 from mixtral_moe_model import Transformer as MixtralMoE
 from mixtral_moe_quantize import (
     WeightOnlyInt8QuantHandler as MixtralMoEWeightOnlyInt8QuantHandler,
@@ -13,9 +15,6 @@ from mixtral_moe_quantize import (
 from model import Transformer as LLaMA
 from quantize import WeightOnlyInt8QuantHandler as LLaMAWeightOnlyInt8QuantHandler
 
-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

@@ -1,10 +1,10 @@
 # flake8: noqa: E266, C417, B950
-from mixtral_moe_model import ConditionalFeedForward
-
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
+from mixtral_moe_model import ConditionalFeedForward
+
 ##### Quantization Primitives ######
 
 

benchmarks/inference/server.py

@@ -187,9 +187,8 @@ class BackendWorker:
     def _setup(self):
         import time
 
-        from torchvision.models.resnet import BasicBlock, ResNet
-
         import torch
+        from torchvision.models.resnet import BasicBlock, ResNet
 
         # Create ResNet18 on meta device
         with torch.device("meta"):

benchmarks/instruction_counts/core/utils.py

@@ -4,11 +4,11 @@ import shutil
 import textwrap
 from typing import List, Optional, Tuple
 
+from torch.utils.benchmark.utils.common import _make_temp_dir
+
 from core.api import GroupedBenchmark, TimerArgs
 from core.types import Definition, FlatIntermediateDefinition, Label
 
-from torch.utils.benchmark.utils.common import _make_temp_dir
-
 
 _TEMPDIR: Optional[str] = None
 

benchmarks/operator_benchmark/benchmark_core.py

@@ -7,7 +7,6 @@ from collections import namedtuple
 
 import benchmark_utils
 import numpy as np
-
 import torch
 
 # needs to be imported after torch

benchmarks/operator_benchmark/benchmark_pytorch.py

@@ -2,7 +2,6 @@ import json
 import time
 
 import benchmark_cpp_extension  # noqa: F401
-
 import torch
 
 

benchmarks/operator_benchmark/common/repeat_benchmark.py

@@ -1,7 +1,6 @@
 import time
 
 import numpy as np
-
 import torch
 
 """Microbenchmarks for Tensor repeat operator. Supports PyTorch."""

benchmarks/operator_benchmark/common/tests/add_ops_list_test.py

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

benchmarks/operator_benchmark/common/tests/jit_forward_test.py

@@ -1,5 +1,4 @@
 import operator_benchmark as op_bench
-
 import torch
 
 intraop_bench_configs = op_bench.config_list(

benchmarks/operator_benchmark/common/tests/pt_backward_test.py

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

benchmarks/operator_benchmark/common/tests/pt_configs_list_test.py

@@ -1,5 +1,4 @@
 import operator_benchmark as op_bench
-
 import torch
 
 """Microbenchmarks for element-wise Add operator. Supports both Caffe2/PyTorch."""

benchmarks/operator_benchmark/common/tests/pt_cpu_gpu_forward_backward_test.py

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

benchmarks/operator_benchmark/common/tests/random_sample_test.py

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

benchmarks/operator_benchmark/pt/add_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for add_ operator. Supports both Caffe2/PyTorch."""
 
 # Configs for PT add operator

benchmarks/operator_benchmark/pt/ao_sparsifier_test.py

@@ -1,9 +1,10 @@
-import operator_benchmark as op_bench
 import torch
 from torch import nn
 
 from torch.ao import pruning
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for sparsifier."""
 

benchmarks/operator_benchmark/pt/as_strided_test.py

@@ -1,9 +1,9 @@
 from typing import List
 
-import operator_benchmark as op_bench
-
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for as_strided operator"""
 

benchmarks/operator_benchmark/pt/batchnorm_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn.functional as F
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for batchnorm operator."""
 

benchmarks/operator_benchmark/pt/binary_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for binary operators."""
 

benchmarks/operator_benchmark/pt/bmm_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for add_ operator. Supports both Caffe2/PyTorch."""
 
 

benchmarks/operator_benchmark/pt/cat_test.py

@@ -1,10 +1,10 @@
 import random
 from typing import List
 
-import operator_benchmark as op_bench
-
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for Cat operator"""
 

benchmarks/operator_benchmark/pt/channel_shuffle_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for channel_shuffle operator."""
 

benchmarks/operator_benchmark/pt/chunk_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for Chunk operator"""
 

benchmarks/operator_benchmark/pt/clip_ranges_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for ClipRanges operator."""
 torch.ops.load_library("//caffe2/torch/fb/sparsenn:sparsenn_operators")

benchmarks/operator_benchmark/pt/conv_test.py

@@ -1,8 +1,9 @@
+import torch
+import torch.nn as nn
+
 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

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for diag operator"""
 

benchmarks/operator_benchmark/pt/embeddingbag_test.py

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

benchmarks/operator_benchmark/pt/gather_test.py

@@ -1,7 +1,7 @@
 import numpy
+import torch
 
 import operator_benchmark as op_bench
-import torch
 
 
 """Microbenchmarks for gather operator."""

benchmarks/operator_benchmark/pt/gelu_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """
 Microbenchmarks for the gelu operators.

benchmarks/operator_benchmark/pt/groupnorm_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn.functional as F
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for groupnorm operator."""
 

benchmarks/operator_benchmark/pt/hardsigmoid_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn as nn
 
+import operator_benchmark as op_bench
+
 
 """
 Microbenchmarks for the hardsigmoid operator.

benchmarks/operator_benchmark/pt/hardswish_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn as nn
 
+import operator_benchmark as op_bench
+
 
 """
 Microbenchmarks for the hardswish operators.

benchmarks/operator_benchmark/pt/index_select_test.py

@@ -1,7 +1,7 @@
 import numpy
+import torch
 
 import operator_benchmark as op_bench
-import torch
 
 
 """Microbenchmarks for index_select operator."""

benchmarks/operator_benchmark/pt/instancenorm_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn.functional as F
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for instancenorm operator."""
 

benchmarks/operator_benchmark/pt/interpolate_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for interpolate operator."""
 
 

benchmarks/operator_benchmark/pt/layernorm_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn.functional as F
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for layernorm operator."""
 

benchmarks/operator_benchmark/pt/linear_prepack_fp16_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for linear_prepack_fp16_ operator. Supports both Caffe2/PyTorch."""
 
 # Configs for PT linear_prepack_fp16 operator

benchmarks/operator_benchmark/pt/linear_test.py

@@ -1,8 +1,9 @@
+import torch
+import torch.nn as nn
+
 from pt import configs
 
 import operator_benchmark as op_bench
-import torch
-import torch.nn as nn
 
 
 """Microbenchmarks for Linear operator."""

benchmarks/operator_benchmark/pt/linear_unpack_fp16_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """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 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for MatMul operator"""
 
 # Configs for PT Matmul operator

benchmarks/operator_benchmark/pt/matrix_mult_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """
 Microbenchmarks for batch matrix mult with einsum and torch.bmm.
 """

benchmarks/operator_benchmark/pt/nan_to_num_test.py

@@ -1,9 +1,9 @@
 import math
 
-import operator_benchmark as op_bench
-
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for torch.nan_to_num / nan_to_num_ operators"""
 

benchmarks/operator_benchmark/pt/pool_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn as nn
 
+import operator_benchmark as op_bench
+
 """
 Microbenchmarks for MaxPool1d and AvgPool1d operators.
 """

benchmarks/operator_benchmark/pt/qactivation_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.ao.nn.quantized.functional as qF
 
+import operator_benchmark as op_bench
+
 r"""Microbenchmarks for the quantized activations."""
 
 qactivation_long_configs = op_bench.cross_product_configs(

benchmarks/operator_benchmark/pt/qarithmetic_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 from torch._ops import ops
 
+import operator_benchmark as op_bench
+
 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

@@ -1,11 +1,11 @@
 import numpy
-from pt import configs
-
-import operator_benchmark as op_bench
 import torch
 import torch.ao.nn.qat as nnqat
+from pt import configs
 from torch.ao.quantization import default_embedding_qat_qconfig
 
+import operator_benchmark as op_bench
+
 """
 Microbenchmarks for QAT Embedding + EmbeddingBag operators.
 """

benchmarks/operator_benchmark/pt/qbatchnorm_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized batchnorm operator."""
 

benchmarks/operator_benchmark/pt/qcat_test.py

@@ -1,10 +1,10 @@
 from typing import List
 
-import operator_benchmark as op_bench
-
 import torch
 import torch.ao.nn.quantized as nnq
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized Cat operator"""
 

benchmarks/operator_benchmark/pt/qcomparators_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 qcomparators_configs = op_bench.cross_product_configs(
     N=(8, 64),
     dtype=(torch.quint8, torch.qint8, torch.qint32),

benchmarks/operator_benchmark/pt/qconv_test.py

@@ -1,8 +1,9 @@
+import torch
+import torch.ao.nn.quantized as nnq
+
 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

@@ -1,12 +1,12 @@
 from typing import Optional
 
 import numpy as np
-
-import operator_benchmark as op_bench
 import torch
 
 from torch.testing._internal.common_quantization import lengths_to_offsets
 
+import operator_benchmark as op_bench
+
 torch.ops.load_library("//caffe2/torch/fb/sparsenn:sparsenn_operators")
 
 

benchmarks/operator_benchmark/pt/qembedding_pack_test.py

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

@@ -1,9 +1,9 @@
 import numpy
+import torch
+import torch.ao.nn.quantized as nnq
 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

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized groupnorm operator."""
 

benchmarks/operator_benchmark/pt/qinstancenorm_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized instancenorm operator."""
 

benchmarks/operator_benchmark/pt/qinterpolate_test.py

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

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized layernorm operator."""
 

benchmarks/operator_benchmark/pt/qlinear_test.py

@@ -1,10 +1,11 @@
-from pt import configs
-
-import operator_benchmark as op_bench
 import torch
 import torch.ao.nn.quantized as nnq
 import torch.ao.nn.quantized.dynamic as nnqd
 
+from pt import configs
+
+import operator_benchmark as op_bench
+
 """
 Microbenchmarks for Quantized Linear operators.
 """

benchmarks/operator_benchmark/pt/qobserver_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.ao.quantization.observer as obs
 
+import operator_benchmark as op_bench
+
 qobserver_short_configs_dict = {
     "attr_names": ("C", "M", "N", "dtype", "device"),
     "attrs": (

benchmarks/operator_benchmark/pt/qpool_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 # 2D pooling will have input matrix of rank 3 or 4
 qpool2d_long_configs = op_bench.config_list(
     attrs=(

benchmarks/operator_benchmark/pt/qrnn_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 from torch import nn
 
+import operator_benchmark as op_bench
+
 """
 Microbenchmarks for RNNs.
 """

benchmarks/operator_benchmark/pt/qtensor_method_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 # Configs for pointwise and reduction unary ops
 qmethods_configs_short = op_bench.config_list(
     attr_names=["M", "N"],

benchmarks/operator_benchmark/pt/quantization_test.py

@@ -1,9 +1,10 @@
-import operator_benchmark as op_bench
 import torch
 import torch.ao.nn.quantized as nnq
 import torch.ao.quantization as tq
 import torch.nn as nn
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for general quantization operations."""
 
 # mode is used to show the direction of the benchmark:

benchmarks/operator_benchmark/pt/qunary_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for quantized unary operators (point-wise and reduction)."""
 

benchmarks/operator_benchmark/pt/remainder_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for remainder operators."""
 

benchmarks/operator_benchmark/pt/softmax_test.py

@@ -1,7 +1,8 @@
-import operator_benchmark as op_bench
 import torch
 import torch.nn as nn
 
+import operator_benchmark as op_bench
+
 
 """
 Microbenchmarks for the softmax operators.

benchmarks/operator_benchmark/pt/split_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for Split operator"""
 

benchmarks/operator_benchmark/pt/stack_test.py

@@ -1,10 +1,10 @@
 import random
 from typing import List
 
-import operator_benchmark as op_bench
-
 import torch
 
+import operator_benchmark as op_bench
+
 
 """Microbenchmarks for Stack operator"""
 

benchmarks/operator_benchmark/pt/sum_test.py

@@ -1,6 +1,7 @@
-import operator_benchmark as op_bench
 import torch
 
+import operator_benchmark as op_bench
+
 """Microbenchmarks for sum reduction operator."""
 
 # Configs for PT add operator