[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

Author: pre-commit-ci[bot]

neural_compressor/adaptor/torch_utils/waq/smooth_quant.py

@@ -363,7 +363,7 @@ def _parse_absorb_to_layers(self, op_types, folding):
                 input_mins, input_maxes = calib.calibrate(
                     1, op_types
                 )  ##TODO if using qfunc for calibration, it will calibrate twice
-        #use qfunc to calibrate, the input min could be used for fixed alpha transformation
+            # use qfunc to calibrate, the input min could be used for fixed alpha transformation
             self.input_mins = input_mins
             self.input_maxes = input_maxes
             diff_modules = set(self.absorb_to_layer.keys()).difference(input_mins.keys())
@@ -434,7 +434,7 @@ def transform(
             for key in self.input_mins.keys():
                 input_maxes_abs[key] = torch.max(torch.abs(self.input_mins[key]), torch.abs(self.input_maxes[key]))
             if self.q_func:
-                self.need_calibration = False # Avoid double-calibration in fixed-value alpha SQ.
+                self.need_calibration = False  # Avoid double-calibration in fixed-value alpha SQ.
 
         if self.absorb_to_layer is None:
             logger.warning("empty absorb_to_layer, smoothquant is ignored ")
@@ -458,7 +458,7 @@ def transform(
             self.alpha = auto_alpha_tuner.tune()
             input_maxes_abs = auto_alpha_tuner.input_maxes_abs
             self.input_mins, self.input_maxes = auto_alpha_tuner.input_mins, auto_alpha_tuner.input_maxes
-            if auto_alpha_tuner.loss_type == 'blockwise':
+            if auto_alpha_tuner.loss_type == "blockwise":
                 self.block_names = auto_alpha_tuner.block_names
 
         elif self.need_calibration:

test/algorithm/test_smooth_quant.py

@@ -1537,12 +1537,20 @@ def forward(self, x):
             "alpha_min": 0.5,
             "alpha_max": 0.9,
             "alpha_step": 0.1,
-            "shared_criterion": 'mean',
+            "shared_criterion": "mean",
             "init_alpha": 0.7,
-            "n_samples": 32
+            "n_samples": 32,
         }
-        alpha_tuner = AutoAlpha(model, sq.dataloader, sq.absorb_to_layer, op_types=[torch.nn.Linear, torch.nn.Conv2d], \
-            device=sq.device, q_func=sq.q_func, example_inputs=sq.example_inputs, **auto_alpha_args)
+        alpha_tuner = AutoAlpha(
+            model,
+            sq.dataloader,
+            sq.absorb_to_layer,
+            op_types=[torch.nn.Linear, torch.nn.Conv2d],
+            device=sq.device,
+            q_func=sq.q_func,
+            example_inputs=sq.example_inputs,
+            **auto_alpha_args,
+        )
         tuned_alpha = alpha_tuner.tune()
         assert alpha_tuner.init_alpha == 0.7
         assert alpha_tuner.alpha_min == 0.5
@@ -1563,12 +1571,7 @@ def test_sq_linear_Blockwise_auto(self):
             "shared_criterion": "mean",
             "do_blockwise": True,
         }
-        sq.transform(
-            alpha="auto",
-            calib_iter=1,
-            folding=False,
-            auto_alpha_args=auto_alpha_args
-        )
+        sq.transform(alpha="auto", calib_iter=1, folding=False, auto_alpha_args=auto_alpha_args)
         for i in range(12):
             op_name1 = "model.decoder.layers." + str(i) + ".self_attn.out_proj"
             op_name2 = "model.decoder.layers." + str(i) + ".fc1"