Update quality tools to 2023 (huggingface#1046)

* Setup 2023 tooling for quality

* Result of styling

* Simplify inits and remove isort and flake8 from doc

* Puts back isort skip flag

.gitignore

@@ -138,4 +138,7 @@ dmypy.json
 .DS_Store
 
 # More test things
-wandb
+wandb
+
+# ruff
+.ruff_cache

CONTRIBUTING.md

@@ -152,7 +152,7 @@ Follow these steps to start contributing:
    $ make test
    ```
 
-   `accelerate` relies on `black` and `isort` to format its source code
+   `accelerate` relies on `black` and `ruff` to format its source code
    consistently. After you make changes, apply automatic style corrections and code verifications
    that can't be automated in one go with:
 
@@ -165,7 +165,7 @@ Follow these steps to start contributing:
    $ make style
    ```
 
-   `accelerate` also uses `flake8` and a few custom scripts to check for coding mistakes. Quality
+   `accelerate` also uses a few custom scripts to check for coding mistakes. Quality
    control runs in CI, however you can also run the same checks with:
 
    ```bash

Makefile

@@ -13,14 +13,13 @@ extra_quality_checks:
 # this target runs checks on all files
 quality:
 	black --check $(check_dirs)
-	isort --check-only $(check_dirs)
-	flake8 $(check_dirs)
+	ruff $(check_dirs)
 	doc-builder style src/accelerate docs/source --max_len 119 --check_only
 
 # Format source code automatically and check is there are any problems left that need manual fixing
 style:
 	black $(check_dirs)
-	isort $(check_dirs)
+	ruff $(check_dirs) --fix
 	doc-builder style src/accelerate docs/source --max_len 119
 
 # Run tests for the library

benchmarks/big_model_inference.py

@@ -16,12 +16,12 @@
 import time
 
 import torch
-
 import transformers
-from accelerate.utils import compute_module_sizes
 from measures_util import end_measure, log_measures, start_measure
 from transformers import AutoConfig, AutoModelForCausalLM, AutoModelForSeq2SeqLM, AutoTokenizer
 
+from accelerate.utils import compute_module_sizes
+
 
 DEFAULT_MODELS = {
     "gpt-j-6b": {"is_causal": True, "model": "sgugger/sharded-gpt-j-6B", "tokenizer": "EleutherAI/gpt-j-6B"},

benchmarks/measures_util.py

@@ -2,9 +2,8 @@
 import threading
 import time
 
-import torch
-
 import psutil
+import torch
 
 
 class PeakCPUMemory:

docs/source/usage_guides/megatron_lm.mdx

@@ -290,6 +290,7 @@ You will implement the `accelerate.utils.AbstractTrainStep` or inherit from thei
 ```python
 from accelerate.utils import MegatronLMDummyScheduler, GPTTrainStep, avg_losses_across_data_parallel_group
 
+
 # Custom loss function for the Megatron model
 class GPTTrainStepWithCustomLoss(GPTTrainStep):
     def __init__(self, megatron_args, **kwargs):

examples/by_feature/automatic_gradient_accumulation.py

@@ -14,16 +14,16 @@
 import argparse
 import os
 
+# New Code #
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-# New Code #
-import evaluate
 from accelerate import Accelerator, DistributedType
 from accelerate.utils import find_executable_batch_size
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/checkpointing.py

@@ -15,14 +15,14 @@
 import argparse
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/cross_validation.py

@@ -15,20 +15,20 @@
 import argparse
 from typing import List
 
+import evaluate
 import numpy as np
 import torch
-from torch.optim import AdamW
-from torch.utils.data import DataLoader
-
-import evaluate
-from accelerate import Accelerator, DistributedType
 from datasets import DatasetDict, load_dataset
 
 # New Code #
 # We'll be using StratifiedKFold for this example
 from sklearn.model_selection import StratifiedKFold
+from torch.optim import AdamW
+from torch.utils.data import DataLoader
 from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
+from accelerate import Accelerator, DistributedType
+
 
 ########################################################################
 # This is a fully working simple example to use Accelerate,

examples/by_feature/deepspeed_with_config_support.py

@@ -31,16 +31,12 @@
 from itertools import chain
 from pathlib import Path
 
-import torch
-from torch.utils.data import DataLoader
-
 import datasets
+import torch
 import transformers
-from accelerate import Accelerator, DistributedType
-from accelerate.logging import get_logger
-from accelerate.utils import DummyOptim, DummyScheduler, set_seed
 from datasets import load_dataset
 from huggingface_hub import Repository
+from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 from transformers import (
     CONFIG_MAPPING,
@@ -55,6 +51,10 @@
 from transformers.utils import get_full_repo_name
 from transformers.utils.versions import require_version
 
+from accelerate import Accelerator, DistributedType
+from accelerate.logging import get_logger
+from accelerate.utils import DummyOptim, DummyScheduler, set_seed
+
 
 logger = get_logger(__name__)
 

examples/by_feature/fsdp_with_peak_mem_tracking.py

@@ -16,13 +16,13 @@
 import gc
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/gradient_accumulation.py

@@ -15,14 +15,14 @@
 import argparse
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/megatron_lm_gpt_pretraining.py

@@ -31,16 +31,12 @@
 from itertools import chain
 from pathlib import Path
 
-import torch
-from torch.utils.data import DataLoader
-
 import datasets
+import torch
 import transformers
-from accelerate import Accelerator, DistributedType
-from accelerate.logging import get_logger
-from accelerate.utils import MegatronLMDummyScheduler, set_seed
 from datasets import load_dataset
 from huggingface_hub import Repository
+from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
 from transformers import (
     CONFIG_MAPPING,
@@ -55,6 +51,10 @@
 from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
+from accelerate import Accelerator, DistributedType
+from accelerate.logging import get_logger
+from accelerate.utils import MegatronLMDummyScheduler, set_seed
+
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
 check_min_version("4.23.0.dev0")

examples/by_feature/memory.py

@@ -14,16 +14,16 @@
 import argparse
 import os
 
+# New Code #
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-# New Code #
-import evaluate
 from accelerate import Accelerator, DistributedType
 from accelerate.utils import find_executable_batch_size
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/multi_process_metrics.py

@@ -15,14 +15,14 @@
 import argparse
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/by_feature/tracking.py

@@ -15,14 +15,14 @@
 import argparse
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/complete_cv_example.py

@@ -17,14 +17,14 @@
 import re
 
 import numpy as np
+import PIL
 import torch
+from timm import create_model
 from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import DataLoader, Dataset
+from torchvision.transforms import Compose, RandomResizedCrop, Resize, ToTensor
 
-import PIL
 from accelerate import Accelerator
-from timm import create_model
-from torchvision.transforms import Compose, RandomResizedCrop, Resize, ToTensor
 
 
 ########################################################################

examples/complete_nlp_example.py

@@ -15,14 +15,14 @@
 import argparse
 import os
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

examples/cv_example.py

@@ -17,14 +17,14 @@
 import re
 
 import numpy as np
+import PIL
 import torch
+from timm import create_model
 from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import DataLoader, Dataset
+from torchvision.transforms import Compose, RandomResizedCrop, Resize, ToTensor
 
-import PIL
 from accelerate import Accelerator
-from timm import create_model
-from torchvision.transforms import Compose, RandomResizedCrop, Resize, ToTensor
 
 
 ########################################################################

examples/nlp_example.py

@@ -14,14 +14,14 @@
 # limitations under the License.
 import argparse
 
+import evaluate
 import torch
+from datasets import load_dataset
 from torch.optim import AdamW
 from torch.utils.data import DataLoader
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
-import evaluate
 from accelerate import Accelerator, DistributedType
-from datasets import load_dataset
-from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
 
 
 ########################################################################

pyproject.toml

@@ -1,3 +1,17 @@
 [tool.black]
 line-length = 119
-target-version = ['py36']
+target-version = ['py37']
+
+[tool.ruff]
+# Never enforce `E501` (line length violations).
+ignore = ["E501", "E741", "W605"]
+select = ["E", "F", "I", "W"]
+line-length = 119
+
+# Ignore import violations in all `__init__.py` files.
+[tool.ruff.per-file-ignores]
+"__init__.py" = ["E402", "F401", "F403", "F811"]
+
+[tool.ruff.isort]
+lines-after-imports = 2
+known-first-party = ["accelerate"]

setup.cfg

@@ -4,11 +4,6 @@ ensure_newline_before_comments = True
 force_grid_wrap = 0
 include_trailing_comma = True
 known_first_party = accelerate
-known_third_party =
-    numpy
-    torch
-    torch_xla
-
 line_length = 119
 lines_after_imports = 2
 multi_line_output = 3