[5/N][Easy] fix typo for usort config in pyproject.toml (kown -…

…> `known`): sort torch (pytorch#127126)

The `usort` config in `pyproject.toml` has no effect due to a typo. Fixing the typo make `usort` do more and generate the changes in the PR. Except `pyproject.toml`, all changes are generated by `lintrunner -a --take UFMT --all-files`.

Pull Request resolved: pytorch#127126
Approved by: https://github.com/kit1980

android/test_app/make_assets.py

@@ -1,6 +1,7 @@
-import torch
 from torchvision import models
 
+import torch
+
 print(torch.version.__version__)
 
 resnet18 = models.resnet18(weights=models.ResNet18_Weights.IMAGENET1K_V1)

android/test_app/make_assets_custom.py

@@ -4,10 +4,11 @@
 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

@@ -18,11 +18,12 @@
 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 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

@@ -3,18 +3,10 @@
 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,
@@ -25,6 +17,14 @@
     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 time
 from abc import ABC, abstractmethod
 
-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
 
+import torch
+import torch.distributed.rpc as rpc
+
 
 class ParameterServerBase(ABC):
     PARAMETER_SERVER_BATCH_METRIC = "parameter_server_batch_metric"

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

@@ -1,6 +1,7 @@
+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 time
 from abc import ABC, abstractmethod
 
-import torch
-
 from metrics.MetricsLogger import MetricsLogger
 
+import torch
+
 
 class TrainerBase(ABC):
     BATCH_LEVEL_METRIC = "batch_level_metric"

benchmarks/distributed/rpc/rl/coordinator.py

@@ -2,12 +2,12 @@
 
 import numpy as np
 
-import torch
-import torch.distributed.rpc as rpc
-
 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

@@ -4,11 +4,11 @@
 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

@@ -36,20 +36,21 @@
     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 (
@@ -59,8 +60,6 @@
     same,
 )
 
-from tqdm.auto import tqdm, trange
-
 try:
     from torch._dynamo.utils import (
         clone_inputs,
@@ -74,15 +73,14 @@
         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
@@ -2343,17 +2341,17 @@ def get_benchmark_indices(self, length):
 
     def get_fsdp_auto_wrap_policy(self, model_name: str):
         from diffusers.models.transformer_2d import Transformer2DModel
-
-        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
 
+        from torch.distributed.fsdp.wrap import (
+            ModuleWrapPolicy,
+            size_based_auto_wrap_policy,
+        )
+
         # handcrafted wrap policy
         MODEL_FSDP_WRAP = {
             "stable_diffusion_unet": (Transformer2DModel,),

benchmarks/dynamo/huggingface.py

@@ -7,9 +7,10 @@
 import sys
 import warnings
 
-import torch
 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
 

benchmarks/dynamo/microbenchmarks/bench_mm_fusion.py

@@ -1,10 +1,11 @@
 # 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

@@ -1,9 +1,10 @@
+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

@@ -1,8 +1,9 @@
 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

@@ -1,10 +1,11 @@
+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
-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

@@ -1,8 +1,9 @@
+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

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

benchmarks/dynamo/microbenchmarks/operatorbench.py

@@ -1,9 +1,10 @@
 #!/usr/bin/env python3
 import click
 import numpy as np
-import torch
 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

benchmarks/dynamo/runner.py

@@ -47,13 +47,14 @@
 
 import numpy as np
 import pandas as pd
-import torch
-
-import torch._dynamo
 from matplotlib import rcParams
 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/timm_models.py

@@ -7,9 +7,10 @@
 import sys
 import warnings
 
-import torch
 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
 

benchmarks/dynamo/torchbench.py

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

benchmarks/dynamo/training_loss.py

@@ -5,12 +5,13 @@
 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