Enable more Ruff lints & fix issues (#2419)

* Remove antiquated flake8 and isort configuration

* Bump to Ruff 0.2.1

* Explain ruff options

* Autofix Ruff B010 (static `setattr`)

* Autofix Ruff B009 (static `getattr`)

* Enable Ruff UP (not UP007); auto-fix

* Fix remaining Ruff UP complaints

* Fix a couple more format calls

docs/source/usage_guides/deepspeed.md

@@ -656,7 +656,7 @@ ZeRO Stage-3 has 2 options:
    Below is the snippet from `examples/by_feature/deepspeed_with_config_support.py` showing this:
    ```python
    success = model.save_checkpoint(PATH, ckpt_id, checkpoint_state_dict)
-   status_msg = "checkpointing: PATH={}, ckpt_id={}".format(PATH, ckpt_id)
+   status_msg = f"checkpointing: PATH={PATH}, ckpt_id={ckpt_id}"
    if success:
        logging.info(f"Success {status_msg}")
    else:

examples/by_feature/checkpointing.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/cross_validation.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/deepspeed_with_config_support.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/early_stopping.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/fsdp_with_peak_mem_tracking.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -334,13 +333,11 @@ def collate_fn(examples):
                         accelerator.save_state(output_dir)
         # New Code #
         # Printing the GPU memory usage details such as allocated memory, peak memory, and total memory usage
-        accelerator.print("Memory before entering the train : {}".format(b2mb(tracemalloc.begin)))
-        accelerator.print("Memory consumed at the end of the train (end-begin): {}".format(tracemalloc.used))
-        accelerator.print("Peak Memory consumed during the train (max-begin): {}".format(tracemalloc.peaked))
+        accelerator.print(f"Memory before entering the train : {b2mb(tracemalloc.begin)}")
+        accelerator.print(f"Memory consumed at the end of the train (end-begin): {tracemalloc.used}")
+        accelerator.print(f"Peak Memory consumed during the train (max-begin): {tracemalloc.peaked}")
         accelerator.print(
-            "Total Peak Memory consumed during the train (max): {}".format(
-                tracemalloc.peaked + b2mb(tracemalloc.begin)
-            )
+            f"Total Peak Memory consumed during the train (max): {tracemalloc.peaked + b2mb(tracemalloc.begin)}"
         )
         # Logging the peak memory usage of the GPU to the tracker
         if args.with_tracking:
@@ -387,11 +384,11 @@ def collate_fn(examples):
                 accelerator.save_state(output_dir)
         # New Code #
         # Printing the GPU memory usage details such as allocated memory, peak memory, and total memory usage
-        accelerator.print("Memory before entering the eval : {}".format(b2mb(tracemalloc.begin)))
-        accelerator.print("Memory consumed at the end of the eval (end-begin): {}".format(tracemalloc.used))
-        accelerator.print("Peak Memory consumed during the eval (max-begin): {}".format(tracemalloc.peaked))
+        accelerator.print(f"Memory before entering the eval : {b2mb(tracemalloc.begin)}")
+        accelerator.print(f"Memory consumed at the end of the eval (end-begin): {tracemalloc.used}")
+        accelerator.print(f"Peak Memory consumed during the eval (max-begin): {tracemalloc.peaked}")
         accelerator.print(
-            "Total Peak Memory consumed during the eval (max): {}".format(tracemalloc.peaked + b2mb(tracemalloc.begin))
+            f"Total Peak Memory consumed during the eval (max): {tracemalloc.peaked + b2mb(tracemalloc.begin)}"
         )
         # Logging the peak memory usage of the GPU to the tracker
         if args.with_tracking:

examples/by_feature/gradient_accumulation.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/local_sgd.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2023 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/megatron_lm_gpt_pretraining.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/multi_process_metrics.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/by_feature/tracking.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/complete_cv_example.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/complete_nlp_example.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/cv_example.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/inference/bert.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/inference/gpt2.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/inference/llama.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/inference/t5.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

examples/nlp_example.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

pyproject.toml

@@ -1,15 +1,32 @@
 [tool.ruff]
-# Never enforce `E501` (line length violations).
-ignore = ["E501", "E741", "W605"]
-select = ["E", "F", "I", "W"]
 line-length = 119
+target-version = "py38"
 
-# Ignore import violations in all `__init__.py` files.
-[tool.ruff.per-file-ignores]
-"__init__.py" = ["E402", "F401", "F403", "F811"]
+[tool.ruff.lint]
+ignore-init-module-imports = true
+extend-select = [
+    "B009", # static getattr
+    "B010", # static setattr
+    "E", # PEP8 errors
+    "F", # PEP8 formatting
+    "I", # Import sorting
+    "W", # PEP8 warnings
+    "UP", # Pyupgrade
+]
+ignore = [
+    "E501", # Line length (handled by ruff-format)
+    "E741", # Ambiguous variable name
+    "W605", # Invalid escape sequence
+    "UP007", # X | Y type annotations
+]
+
+[tool.ruff.lint.per-file-ignores]
+"__init__.py" = [
+    "F401", # Ignore seemingly unused imports (they're meant for re-export)
+]
 "manim_animations/*" = ["ALL"]
 
-[tool.ruff.isort]
+[tool.ruff.lint.isort]
 lines-after-imports = 2
 known-first-party = ["accelerate"]
 

setup.py

@@ -19,7 +19,7 @@
 extras["quality"] = [
     "black ~= 23.1",  # hf-doc-builder has a hidden dependency on `black`
     "hf-doc-builder >= 0.3.0",
-    "ruff ~= 0.1.15",
+    "ruff ~= 0.2.1",
 ]
 extras["docs"] = []
 extras["test_prod"] = ["pytest", "pytest-xdist", "pytest-subtests", "parameterized"]
@@ -49,7 +49,7 @@
     name="accelerate",
     version="0.28.0.dev",
     description="Accelerate",
-    long_description=open("README.md", "r", encoding="utf-8").read(),
+    long_description=open("README.md", encoding="utf-8").read(),
     long_description_content_type="text/markdown",
     keywords="deep learning",
     license="Apache",

src/accelerate/accelerator.py

@@ -430,15 +430,14 @@ def __init__(
         # Mixed precision attributes
         self.scaler = None
         self.native_amp = False
-        err = "{mode} mixed precision requires {requirement}"
         if (
             self.state.mixed_precision == "fp16"
             and self.device.type != "cpu"
             and self.distributed_type not in (DistributedType.DEEPSPEED, DistributedType.MEGATRON_LM)
         ):
             self.native_amp = True
             if self.device.type not in ("xpu", "cuda", "mps", "npu"):
-                raise ValueError(err.format(mode="fp16", requirement="a GPU"))
+                raise ValueError(f"fp16 mixed precision requires a GPU (not {self.device.type!r}).")
             kwargs = self.scaler_handler.to_kwargs() if self.scaler_handler is not None else {}
             if self.distributed_type == DistributedType.FSDP:
                 from torch.distributed.fsdp.sharded_grad_scaler import ShardedGradScaler
@@ -458,7 +457,7 @@ def __init__(
             else:
                 self.native_amp = is_bf16_available(True)
             if mixed_precision == "bf16" and not self.native_amp and not is_tpu_available():
-                raise ValueError(err.format(mode="bf16", requirement="PyTorch >= 1.10 and a supported device."))
+                raise ValueError("bf16 mixed precision requires PyTorch >= 1.10 and a supported device.")
 
         # Start of internal step tracking
         self.step = 0
@@ -1245,7 +1244,7 @@ def prepare(self, *args, device_placement=None):
                 item in container
                 for container in (self._dataloaders, self._models, self._optimizers, self._schedulers)
             ):
-                setattr(item, "_is_accelerate_prepared", True)
+                item._is_accelerate_prepared = True
 
         return result if len(result) > 1 else result[0]
 
@@ -2386,7 +2385,7 @@ def init_trackers(self, project_name: str, config: dict | None = None, init_kwar
                 self.trackers.append(tracker)
             else:
                 tracker_init = LOGGER_TYPE_TO_CLASS[str(tracker)]
-                if getattr(tracker_init, "requires_logging_directory"):
+                if tracker_init.requires_logging_directory:
                     # We can skip this check since it was done in `__init__`
                     self.trackers.append(
                         tracker_init(project_name, self.logging_dir, **init_kwargs.get(str(tracker), {}))

src/accelerate/commands/config/config_args.py

@@ -51,7 +51,7 @@ def load_config_from_file(config_file):
             )
     else:
         config_file = default_config_file
-    with open(config_file, "r", encoding="utf-8") as f:
+    with open(config_file, encoding="utf-8") as f:
         if config_file.endswith(".json"):
             if (
                 json.load(f).get("compute_environment", ComputeEnvironment.LOCAL_MACHINE)
@@ -94,7 +94,7 @@ def to_dict(self):
     @classmethod
     def from_json_file(cls, json_file=None):
         json_file = default_json_config_file if json_file is None else json_file
-        with open(json_file, "r", encoding="utf-8") as f:
+        with open(json_file, encoding="utf-8") as f:
             config_dict = json.load(f)
         if "compute_environment" not in config_dict:
             config_dict["compute_environment"] = ComputeEnvironment.LOCAL_MACHINE
@@ -126,7 +126,7 @@ def to_json_file(self, json_file):
     @classmethod
     def from_yaml_file(cls, yaml_file=None):
         yaml_file = default_yaml_config_file if yaml_file is None else yaml_file
-        with open(yaml_file, "r", encoding="utf-8") as f:
+        with open(yaml_file, encoding="utf-8") as f:
             config_dict = yaml.safe_load(f)
         if "compute_environment" not in config_dict:
             config_dict["compute_environment"] = ComputeEnvironment.LOCAL_MACHINE

src/accelerate/commands/launch.py

@@ -105,19 +105,19 @@ def __call__(self, parser, namespace, values, option_string=None):
             for i, arg in enumerate(opts):
                 # If the argument's container is outside of the used titles, hide it
                 if arg.container.title not in titles + used_titles:
-                    setattr(opts[i], "help", argparse.SUPPRESS)
+                    opts[i].help = argparse.SUPPRESS
                 # If the argument is hardware selection, but not being passed, hide it
                 elif arg.container.title == "Hardware Selection Arguments":
                     if set(arg.option_strings).isdisjoint(set(args)):
-                        setattr(opts[i], "help", argparse.SUPPRESS)
+                        opts[i].help = argparse.SUPPRESS
                     else:
-                        setattr(opts[i], "help", arg.help + " (currently selected)")
+                        opts[i].help = arg.help + " (currently selected)"
                 # If the argument is a training paradigm, but not being passed, hide it
                 elif arg.container.title == "Training Paradigm Arguments":
                     if set(arg.option_strings).isdisjoint(set(used_platforms)):
-                        setattr(opts[i], "help", argparse.SUPPRESS)
+                        opts[i].help = argparse.SUPPRESS
                     else:
-                        setattr(opts[i], "help", arg.help + " (currently selected)")
+                        opts[i].help = arg.help + " (currently selected)"
             for i, group in enumerate(list(parser._action_groups)):
                 # If all arguments in the group are hidden, hide the group
                 if all([arg.help == argparse.SUPPRESS for arg in group._group_actions]):
@@ -924,14 +924,12 @@ def _validate_launch_command(args):
                 args.mixed_precision = defaults.mixed_precision
                 mp_from_config_flag = True
         else:
-            native_amp = False
-            err = "{mode} mixed precision requires {requirement}"
             if args.use_cpu or (args.use_xpu and torch.xpu.is_available()):
                 native_amp = is_torch_version(">=", "1.10")
             else:
                 native_amp = is_bf16_available(True)
             if args.mixed_precision == "bf16" and not native_amp and not (args.tpu and is_tpu_available()):
-                raise ValueError(err.format(mode="bf16", requirement="PyTorch >= 1.10 and a supported device."))
+                raise ValueError("bf16 mixed precision requires PyTorch >= 1.10 and a supported device.")
 
         # Silently set the default here
         if args.dynamo_backend is None:

src/accelerate/commands/menu/input.py

@@ -30,7 +30,7 @@ def mark(key: str):
     def decorator(func):
         handle = getattr(func, "handle_key", [])
         handle += [key]
-        setattr(func, "handle_key", handle)
+        func.handle_key = handle
         return func
 
     return decorator
@@ -44,7 +44,7 @@ def mark_multiple(*keys: List[str]):
     def decorator(func):
         handle = getattr(func, "handle_key", [])
         handle += keys
-        setattr(func, "handle_key", handle)
+        func.handle_key = handle
         return func
 
     return decorator
@@ -58,8 +58,8 @@ class KeyHandler(type):
     def __new__(cls, name, bases, attrs):
         new_cls = super().__new__(cls, name, bases, attrs)
         if not hasattr(new_cls, "key_handler"):
-            setattr(new_cls, "key_handler", {})
-        setattr(new_cls, "handle_input", KeyHandler.handle_input)
+            new_cls.key_handler = {}
+        new_cls.handle_input = KeyHandler.handle_input
 
         for value in attrs.values():
             handled_keys = getattr(value, "handle_key", [])

src/accelerate/commands/tpu.py

@@ -112,7 +112,7 @@ def tpu_command_launcher(args):
         raise ValueError("You must specify either a command file or a command to run on the pod.")
 
     if args.command_file:
-        with open(args.command_file, "r") as f:
+        with open(args.command_file) as f:
             args.command = [f.read().splitlines()]
 
     # To turn list of lists into list of strings

src/accelerate/test_utils/examples.py

@@ -84,14 +84,14 @@ def compare_against_test(base_filename: str, feature_filename: str, parser_only:
             functionalities off of "examples/nlp_example.py", so if `base_filename` is a script other than
             `complete_nlp_example.py`, the template script should be included here. Such as `examples/cv_example.py`
     """
-    with open(base_filename, "r") as f:
+    with open(base_filename) as f:
         base_file_contents = f.readlines()
-    with open(os.path.abspath(os.path.join("examples", "nlp_example.py")), "r") as f:
+    with open(os.path.abspath(os.path.join("examples", "nlp_example.py"))) as f:
         full_file_contents = f.readlines()
-    with open(feature_filename, "r") as f:
+    with open(feature_filename) as f:
         feature_file_contents = f.readlines()
     if secondary_filename is not None:
-        with open(secondary_filename, "r") as f:
+        with open(secondary_filename) as f:
             secondary_file_contents = f.readlines()
 
     # This is our base, we remove all the code from here in our `full_filename` and `feature_filename` to find the new content

src/accelerate/test_utils/scripts/external_deps/test_checkpointing.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -182,7 +181,7 @@ def training_function(config, args):
         accelerator.print("resumed checkpoint performance:", accuracy)
         accelerator.print("resumed checkpoint's scheduler's lr:", lr_scheduler.get_lr()[0])
         accelerator.print("resumed optimizers's lr:", optimizer.param_groups[0]["lr"])
-        with open(os.path.join(args.output_dir, f"state_{starting_epoch-1}.json"), "r") as f:
+        with open(os.path.join(args.output_dir, f"state_{starting_epoch-1}.json")) as f:
             resumed_state = json.load(f)
             assert resumed_state["accuracy"] == accuracy, "Accuracy mismatch, loading from checkpoint failed"
             assert (