Add Ruff for lint and code formatting

Replaces previous shell script which called formatting using `precommit"
command. Ruff ia a tool which manages different formatters and also flake8 linting.

Signed-off-by: Martin Hickey <[email protected]>

Author: hickeyma

.github/workflows/lint.yml

@@ -40,9 +40,9 @@ jobs:
       fail-fast: false
       matrix:
         lint:
-          - name: "fmt"
+          - name: "ruff"
             commands: |
-              tox -e fmt
+              tox -e ruff
           - name: "pylint"
             commands: |
               echo "::add-matcher::.github/workflows/matchers/pylint.json"

.isort.cfg

@@ -1,11 +1,7 @@
 [settings]
-profile=black
-from_first=true
 import_heading_future=Future
 import_heading_stdlib=Standard
 import_heading_thirdparty=Third Party
 import_heading_firstparty=First Party
 import_heading_localfolder=Local
-known_firstparty=
-known_localfolder=fms_mo,tests
 extend_skip=fms_mo/_version.py

.pre-commit-config.yaml

@@ -1,11 +1,15 @@
 repos:
-    - repo: https://github.com/psf/black
-      rev: 22.3.0
-      hooks:
-          - id: black
-            exclude: imports
     - repo: https://github.com/PyCQA/isort
       rev: 5.11.5
       hooks:
           - id: isort
             exclude: imports
+    - repo: https://github.com/astral-sh/ruff-pre-commit
+      # Ruff version.
+      rev: v0.5.0
+      hooks:
+        # Run the linter (most fixers are disabled for now).
+        - id: ruff
+        # Run the formatter.
+        - id: ruff-format
+

examples/PTQ_INT8/run_qa_no_trainer_ptq.py

@@ -19,43 +19,35 @@
 # You can also adapt this script on your own question answering task. Pointers for this are left as comments.
 
 # Standard
-from pathlib import Path
 import argparse
 import json
 import logging
 import math
 import os
 import random
 import time
+from pathlib import Path
 
 # Third Party
+import datasets
+import evaluate
+import numpy as np
+import torch
+import transformers
 from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
 from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
-from transformers import (
-    CONFIG_MAPPING,
-    MODEL_MAPPING,
-    AutoConfig,
-    AutoModelForQuestionAnswering,
-    AutoTokenizer,
-    DataCollatorWithPadding,
-    EvalPrediction,
-    SchedulerType,
-    default_data_collator,
-    get_scheduler,
-)
+from transformers import (CONFIG_MAPPING, MODEL_MAPPING, AutoConfig,
+                          AutoModelForQuestionAnswering, AutoTokenizer,
+                          DataCollatorWithPadding, EvalPrediction,
+                          SchedulerType, default_data_collator, get_scheduler)
 from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions
-import datasets
-import evaluate
-import numpy as np
-import torch
-import transformers
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
 check_min_version("4.39.0.dev0")
@@ -1122,11 +1114,10 @@ def squad_eval(model, keep_model_in_eval_mode=True):
         return eval_metric
 
     # ---- [fms_mo] the following code are added for qat/ptq ----
-    # Local
+    # First Party
     from fms_mo import qconfig_init, qmodel_prep
 
     if args.do_qat:
-
         # create a config dict, if same item exists in both recipe and args, args has the priority.
         qcfg = qconfig_init(recipe="qat_int8", args=args)
 
@@ -1141,8 +1132,7 @@ def squad_eval(model, keep_model_in_eval_mode=True):
         qmodel_prep(model, exam_inp, qcfg, optimizer, use_dynamo=True)
 
     if args.do_ptq:
-
-        # Local
+        # First Party
         from fms_mo.quant.ptq import calib_PTQ_lm
 
         # create a config dict, if same item exists in both recipe and args, args has the priority.
@@ -1177,10 +1167,10 @@ def squad_eval(model, keep_model_in_eval_mode=True):
         from copy import deepcopy
 
         # Third Party
-        from torch.ao.quantization.utils import _parent_name
         import pandas as pd
+        from torch.ao.quantization.utils import _parent_name
 
-        # Local
+        # First Party
         from fms_mo.modules.linear import QLinear, QLinearINT8Deploy
 
         def speedtest(model, exam_inp, Ntest=100):
@@ -1216,17 +1206,16 @@ def speedtest(model, exam_inp, Ntest=100):
             ("int8", "ind"),
             ("int8", "cugr"),
         ]:
-
             logger.info(
                 f"\n    {label} {'with' if comp_mode else 'without'} torch.compile"
             )
             model_copy = deepcopy(model)
 
             if label == "int8":
                 qcfg = qconfig_init(recipe="ptq_int8", args=args)
-                qcfg[
-                    "qmodel_calibration"
-                ] = 0  # no need to run calibration or trained scales will be lost.
+                qcfg["qmodel_calibration"] = (
+                    0  # no need to run calibration or trained scales will be lost.
+                )
                 qmodel_prep(
                     model_copy,
                     exam_inp,
@@ -1479,9 +1468,9 @@ def speedtest(model, exam_inp, Ntest=100):
             "step": completed_steps,
         }
     if args.do_predict:
-        log[
-            "squad_v2_predict" if args.version_2_with_negative else "squad_predict"
-        ] = predict_metric
+        log["squad_v2_predict" if args.version_2_with_negative else "squad_predict"] = (
+            predict_metric
+        )
 
         accelerator.log(log, step=completed_steps)
 

examples/PTQ_INT8/utils_qa.py

@@ -15,16 +15,17 @@
 """
 Post-processing utilities for question answering.
 """
+
 # Standard
-from typing import Optional, Tuple
 import collections
 import json
 import logging
 import os
+from typing import Optional, Tuple
 
 # Third Party
-from tqdm.auto import tqdm
 import numpy as np
+from tqdm.auto import tqdm
 
 logger = logging.getLogger(__name__)
 

examples/QAT_INT8/run_qa_no_trainer_qat.py

@@ -19,42 +19,34 @@
 # You can also adapt this script on your own question answering task. Pointers for this are left as comments.
 
 # Standard
-from pathlib import Path
 import argparse
 import json
 import logging
 import math
 import os
 import random
+from pathlib import Path
 
 # Third Party
+import datasets
+import evaluate
+import numpy as np
+import torch
+import transformers
 from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
 from huggingface_hub import HfApi
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
-from transformers import (
-    CONFIG_MAPPING,
-    MODEL_MAPPING,
-    AutoConfig,
-    AutoModelForQuestionAnswering,
-    AutoTokenizer,
-    DataCollatorWithPadding,
-    EvalPrediction,
-    SchedulerType,
-    default_data_collator,
-    get_scheduler,
-)
+from transformers import (CONFIG_MAPPING, MODEL_MAPPING, AutoConfig,
+                          AutoModelForQuestionAnswering, AutoTokenizer,
+                          DataCollatorWithPadding, EvalPrediction,
+                          SchedulerType, default_data_collator, get_scheduler)
 from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions
-import datasets
-import evaluate
-import numpy as np
-import torch
-import transformers
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
 check_min_version("4.39.0.dev0")
@@ -1068,11 +1060,10 @@ def squad_eval(model, keep_model_in_eval_mode=True):
         return eval_metric
 
     # ---- [fms_mo] the following code are added for qat/ptq ----
-    # Local
+    # First Party
     from fms_mo import qconfig_init, qmodel_prep
 
     if args.do_qat:
-
         # create a config dict, if same item exists in both recipe and args, args has the priority.
         qcfg = qconfig_init(recipe="qat_int8", args=args)
 
@@ -1089,14 +1080,14 @@ def squad_eval(model, keep_model_in_eval_mode=True):
     # ---- [fms_mo] the following code are performing speed tests ----
     elif args.do_lowering:
         # Standard
-        from copy import deepcopy
         import time
+        from copy import deepcopy
 
         # Third Party
-        from torch.ao.quantization.utils import _parent_name
         import pandas as pd
+        from torch.ao.quantization.utils import _parent_name
 
-        # Local
+        # First Party
         from fms_mo.modules.linear import QLinear, QLinearINT8Deploy
 
         def speedtest(model, exam_inp, Ntest=100):
@@ -1132,17 +1123,16 @@ def speedtest(model, exam_inp, Ntest=100):
             ("int8", "ind"),
             ("int8", "cugr"),
         ]:
-
             logger.info(
                 f"\n    {label} {'with' if comp_mode else 'without'} torch.compile"
             )
             model_copy = deepcopy(model)
 
             if label == "int8":
                 qcfg = qconfig_init(recipe="qat_int8", args=args)
-                qcfg[
-                    "qmodel_calibration"
-                ] = 0  # no need to run calibration or trained scales will be lost.
+                qcfg["qmodel_calibration"] = (
+                    0  # no need to run calibration or trained scales will be lost.
+                )
                 qmodel_prep(
                     model_copy,
                     exam_inp,
@@ -1395,9 +1385,9 @@ def speedtest(model, exam_inp, Ntest=100):
             "step": completed_steps,
         }
     if args.do_predict:
-        log[
-            "squad_v2_predict" if args.version_2_with_negative else "squad_predict"
-        ] = predict_metric
+        log["squad_v2_predict" if args.version_2_with_negative else "squad_predict"] = (
+            predict_metric
+        )
 
         accelerator.log(log, step=completed_steps)
 

examples/QAT_INT8/utils_qa.py

@@ -15,16 +15,17 @@
 """
 Post-processing utilities for question answering.
 """
+
 # Standard
-from typing import Optional, Tuple
 import collections
 import json
 import logging
 import os
+from typing import Optional, Tuple
 
 # Third Party
-from tqdm.auto import tqdm
 import numpy as np
+from tqdm.auto import tqdm
 
 logger = logging.getLogger(__name__)
 

fms_mo/__init__.py

@@ -11,14 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""FMS Model Optimizer init. Import most commonly used functions and classes here.
-"""
+"""FMS Model Optimizer init. Import most commonly used functions and classes here."""
 
 # Standard
-from importlib.metadata import PackageNotFoundError, version
 import logging
+from importlib.metadata import PackageNotFoundError, version
 
-# Local
+# First Party
 from fms_mo.prep import qmodel_prep
 from fms_mo.utils.qconfig_utils import qconfig_init
 

fms_mo/calib.py

@@ -11,22 +11,20 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""Main user interfacing functions, such as qmodel_prep()
-
-"""
+"""Main user interfacing functions, such as qmodel_prep()"""
 
 # Standard
-from copy import deepcopy
-from typing import Callable, Tuple, Union
 import logging
 import sys
+from copy import deepcopy
+from typing import Callable, Tuple, Union
 
 # Third Party
+import torch
 from torch import nn
 from transformers.tokenization_utils_base import BatchEncoding
-import torch
 
-# Local
+# First Party
 from fms_mo.modules import QBmm, QConv2d, QConvTranspose2d, QLinear
 from fms_mo.utils.utils import prepare_data_4_fwd, prepare_inputs
 
@@ -186,8 +184,9 @@ def __call__(self, module, inputs):
                 for act, name in [(x, "input"), (hid[layer], "hidden")]:
                     nelem = act.nelement()
                     if self.a_init_method == "percentile":
-                        lower_k, upper_k = int(self.per[0] * nelem), int(
-                            self.per[1] * nelem
+                        lower_k, upper_k = (
+                            int(self.per[0] * nelem),
+                            int(self.per[1] * nelem),
                         )
                         lower_per_cur = (
                             act.reshape(1, -1).kthvalue(lower_k).values.data[0]