Merge pull request #30 from hickeyma/ci/add-ruff-tool

ci: Add Ruff for lint and code formatting

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

@@ -6,6 +6,5 @@ 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

@@ -1126,7 +1126,6 @@ def squad_eval(model, keep_model_in_eval_mode=True):
     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)
 
@@ -1141,7 +1140,6 @@ 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
         from fms_mo.quant.ptq import calib_PTQ_lm
 
@@ -1216,17 +1214,16 @@ 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 +1476,9 @@ 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

@@ -15,6 +15,7 @@
 """
 Post-processing utilities for question answering.
 """
+
 # Standard
 from typing import Optional, Tuple
 import collections

examples/QAT_INT8/run_qa_no_trainer_qat.py

@@ -1072,7 +1072,6 @@ def squad_eval(model, keep_model_in_eval_mode=True):
     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)
 
@@ -1132,17 +1131,16 @@ 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 +1393,9 @@ 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

@@ -15,6 +15,7 @@
 """
 Post-processing utilities for question answering.
 """
+
 # Standard
 from typing import Optional, Tuple
 import collections

fms_mo/__init__.py

@@ -11,8 +11,7 @@
 # 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

fms_mo/calib.py

@@ -11,9 +11,7 @@
 # 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
@@ -204,8 +202,9 @@ 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]

fms_mo/custom_ext_kernels/utils.py

@@ -514,7 +514,6 @@ def exllama_ops_load_and_reg(qcfg=None, run_unit_test=False):
         qcfg["AUTOGPTQ_AVAILABLE"] = True
         need_registration = False
     else:
-
         need_registration = (
             available_packages["exllama_kernels"]
             and available_packages["exllamav2_kernels"]
@@ -597,8 +596,7 @@ def exv2_i4f16_fxinputs(
             q4: int,
             q4_width: int,
             force_cuda: bool,
-        ) -> torch.Tensor:
-            ...
+        ) -> torch.Tensor: ...
 
         # Generic implementation
         @kernel_impl(f"{namespace}::exv2_i4f16_fxinputs", "cuda")