[reland][bc-breaking][quant][be] Refactor fuser_method to include is_qat argument" (#71956)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/71956

Pull Request resolved: https://github.com/facebookresearch/mobile-vision/pull/59

Original commit changeset: f3912e210e8c

Original Phabricator Diff: D33178977 (ef501e8fed)

Test Plan:
Please see original diff for test plans

**Static Docs Preview: classyvision**
|[Full Site](https://our.intern.facebook.com/intern/staticdocs/eph/D33833203/V3/classyvision/)|

|**Modified Pages**|

Reviewed By: andrewor14

Differential Revision: D33833203

fbshipit-source-id: 74a8f22730b00aafa6a173b208e635c1d696959e
(cherry picked from commit fb88772b18)

test/quantization/core/test_workflow_module.py

@@ -820,7 +820,7 @@ class TestDistributed(QuantizationTestCase):
                 torch.ao.quantization.DeQuantStub(),
             )
 
-            torch.ao.quantization.fuse_modules(model, [['1', '2', '3'], ['4', '5']], inplace=True)
+            torch.ao.quantization.fuse_modules_qat(model, [['1', '2', '3'], ['4', '5']], inplace=True)
 
             model.qconfig = torch.ao.quantization.get_default_qat_qconfig('fbgemm')
             torch.ao.quantization.prepare_qat(model, inplace=True)
@@ -861,7 +861,7 @@ class TestDistributed(QuantizationTestCase):
 
             model = Model()
             # fuse it
-            fused_model = torch.ao.quantization.fuse_modules(
+            fused_model = torch.ao.quantization.fuse_modules_qat(
                 model,
                 [['conv', 'bn']],
             )

test/quantization/eager/test_fuse_eager.py

@@ -15,6 +15,7 @@ from torch.ao.quantization import (
     prepare_qat,
     quantize_qat,
     fuse_modules,
+    fuse_modules_qat,
     QConfig,
     default_qconfig,
     default_qat_qconfig,
@@ -43,8 +44,8 @@ class TestFuseEager(QuantizationTestCase):
     def test_fuse_module_train(self):
         model = ModelForFusion(default_qat_qconfig).train()
         # Test step by step fusion
-        model = fuse_modules(model, ['conv1', 'bn1', 'relu1'])
-        model = fuse_modules(model, ['sub1.conv', 'sub1.bn'])
+        model = fuse_modules_qat(model, ['conv1', 'bn1', 'relu1'])
+        model = fuse_modules_qat(model, ['sub1.conv', 'sub1.bn'])
         self.assertEqual(type(model.conv1), nni.ConvBnReLU2d,
                          msg="Fused Conv + BN + Relu first layer")
         self.assertEqual(type(model.bn1), torch.nn.Identity,
@@ -91,7 +92,9 @@ class TestFuseEager(QuantizationTestCase):
             checkQuantized(model)
 
         model = ModelForFusion(default_qat_qconfig).train()
-        model = fuse_modules(model, [['conv1', 'bn1', 'relu1'],
+        model = fuse_modules_qat(
+            model,
+            [['conv1', 'bn1', 'relu1'],
              ['sub1.conv', 'sub1.bn']])
         model = quantize_qat(model, test_only_train_fn, [self.img_data_1d_train])
         with self.assertRaisesRegex(RuntimeError, "Could not run 'aten::native_batch_norm' with arguments from the 'QuantizedCPU'"):
@@ -101,7 +104,9 @@ class TestFuseEager(QuantizationTestCase):
     def test_fuse_module_eval(self):
         model = ModelForFusion(default_qconfig)
         model.eval()
-        model = fuse_modules(model, [['conv3', 'bn3', 'relu4'],
+        model = fuse_modules(
+            model,
+            [['conv3', 'bn3', 'relu4'],
              ['conv1', 'bn1', 'relu1'],
              ['conv2', 'relu2'],
              ['bn2', 'relu3'],
@@ -168,7 +173,9 @@ class TestFuseEager(QuantizationTestCase):
         checkQuantized(model)
 
         model = ModelForFusion(default_qconfig).eval()
-        model = fuse_modules(model, [['conv1', 'bn1', 'relu1'],
+        model = fuse_modules(
+            model,
+            [['conv1', 'bn1', 'relu1'],
              ['conv2', 'relu2'],
              ['bn2', 'relu3'],
              ['sub1.conv', 'sub1.bn'],
@@ -181,11 +188,13 @@ class TestFuseEager(QuantizationTestCase):
             with override_quantized_engine(qengine):
                 model = ModelWithSequentialFusion().train()
                 model.to(torch.float)
-                fuse_modules(model, [['conv1', 'relu1'] ,
+                fuse_modules_qat(
+                    model, [['conv1', 'relu1'] ,
                             ['features.0.0', 'features.0.1', 'features.0.2'],
                             ['features.1.0', 'features.1.1', 'features.1.2'],
                             ['features.2.0', 'features.2.1', 'features.2.2'],
-                                     ['classifier.0', 'classifier.1']], inplace=True)
+                            ['classifier.0', 'classifier.1']],
+                    inplace=True)
                 self.assertEqual(type(model.conv1), nni.ConvReLU2d,
                                  msg="Fused Conv + Relu: nni.ConvReLU2d")
                 self.assertEqual(type(model.conv1[0]), nn.Conv2d,
@@ -233,11 +242,14 @@ class TestFuseEager(QuantizationTestCase):
             with override_quantized_engine(qengine):
                 model = ModelWithSequentialFusion().eval()
                 model.to(torch.float)
-                fuse_modules(model, [['conv1', 'relu1'] ,
+                fuse_modules(
+                    model,
+                    [['conv1', 'relu1'],
                      ['features.0.0', 'features.0.1', 'features.0.2'],
                      ['features.1.0', 'features.1.1', 'features.1.2'],
                      ['features.2.0', 'features.2.1', 'features.2.2'],
-                                     ['classifier.0', 'classifier.1']], inplace=True)
+                     ['classifier.0', 'classifier.1']],
+                    inplace=True)
                 self.assertEqual(type(model.conv1), nni.ConvReLU2d,
                                  msg="Fused Conv + Relu: nni.ConvReLU2d")
                 self.assertEqual(type(model.conv1[0]), nn.Conv2d,
@@ -286,7 +298,9 @@ class TestFuseEager(QuantizationTestCase):
                 # fused model
                 model_orig.qconfig = QConfig(activation=torch.nn.Identity,
                                              weight=torch.nn.Identity)
-                model = fuse_modules(model_orig, [["conv1", "bn1", "relu1"],
+                model = fuse_modules_qat(
+                    model_orig,
+                    [["conv1", "bn1", "relu1"],
                      ["conv2", "bn2"]])
                 prep_model = prepare_qat(model, inplace=False)
                 # output with fusion but no observers.
@@ -385,8 +399,8 @@ class TestFuseEager(QuantizationTestCase):
         self.assertEqual(counter['pre_forwards'], 2 * len(self.img_data_1d))
         self.assertEqual(counter['forwards'], 2 * len(self.img_data_1d))
 
-        model = fuse_modules(model, ['conv1', 'bn1', 'relu1'])
-        model = fuse_modules(model, ['sub1.conv', 'sub1.bn'])
+        model = fuse_modules_qat(model, ['conv1', 'bn1', 'relu1'])
+        model = fuse_modules_qat(model, ['sub1.conv', 'sub1.bn'])
 
         fused = True
         before_fusion_pre_count = counter['pre_forwards']

test/quantization/eager/test_model_numerics.py

@@ -69,7 +69,7 @@ class TestModelNumericsEager(QuantizationTestCase):
                 fq_model = torch.ao.quantization.QuantWrapper(my_model)
                 fq_model.train()
                 fq_model.qconfig = torch.ao.quantization.default_qat_qconfig
-                torch.ao.quantization.fuse_modules(fq_model.module, [['conv1', 'bn1', 'relu1']], inplace=True)
+                torch.ao.quantization.fuse_modules_qat(fq_model.module, [['conv1', 'bn1', 'relu1']], inplace=True)
                 torch.ao.quantization.prepare_qat(fq_model)
                 fq_model.eval()
                 fq_model.apply(torch.ao.quantization.disable_fake_quant)
@@ -105,7 +105,7 @@ class TestModelNumericsEager(QuantizationTestCase):
                     fq_model = torch.ao.quantization.QuantWrapper(my_model)
                     fq_model.train()
                     fq_model.qconfig = qconfig
-                    torch.ao.quantization.fuse_modules(fq_model.module, [['conv1', 'bn1', 'relu1']], inplace=True)
+                    torch.ao.quantization.fuse_modules_qat(fq_model.module, [['conv1', 'bn1', 'relu1']], inplace=True)
                     torch.ao.quantization.prepare_qat(fq_model)
                     fq_model.eval()
                     fq_model.apply(torch.ao.quantization.disable_fake_quant)

test/quantization/fx/test_quantize_fx.py

@@ -48,6 +48,7 @@ from torch.ao.quantization import (
     get_default_qat_qconfig,
     get_default_qconfig_dict,
     fuse_modules,
+    fuse_modules_qat,
     prepare,
     prepare_qat,
     convert,
@@ -363,6 +364,8 @@ class TestFuseFx(QuantizationTestCase):
 
     @skipIfNoFBGEMM
     def test_qconfig_fused_module(self):
+        """ TODO: add test for all fused modules
+        """
         qconfig_dict = {
             "": None,
             "object_type": [(nn.Linear, default_qconfig),
@@ -890,14 +893,19 @@ class TestQuantizeFx(QuantizationTestCase):
                 m_eager.eval()
                 qconfig = get_default_qconfig(qengine)
                 prepare_fn = prepare
+                is_qat = False
             else:
                 m_eager.train()
                 qconfig = get_default_qat_qconfig(qengine)
                 prepare_fn = prepare_qat
+                is_qat = True
 
             fuse_list = ["conv", "bn"]
             if has_relu:
                 fuse_list.append("relu")
+            if is_qat:
+                fuse_modules_qat(m_eager, fuse_list, inplace=True)
+            else:
                 fuse_modules(m_eager, fuse_list, inplace=True)
             m_eager.qconfig = qconfig
             m_eager = prepare_fn(m_eager)
@@ -5847,6 +5855,7 @@ class TestQuantizeFxModels(QuantizationTestCase):
             graph.eval()
             calibrate_or_train = test_only_eval_fn
             data = self.img_data_2d
+            is_qat = False
         else:
             assert quant_type == QuantType.QAT
             qconfig = default_qat_qconfig
@@ -5856,6 +5865,7 @@ class TestQuantizeFxModels(QuantizationTestCase):
             graph.train()
             calibrate_or_train = test_only_train_fn
             data = self.img_data_2d_train
+            is_qat = True
 
         if hasattr(eager, "fuse_model"):
             eager.fuse_model()

torch/ao/quantization/__init__.py

@@ -2,6 +2,7 @@
 
 from .fake_quantize import *  # noqa: F403
 from .fuse_modules import fuse_modules  # noqa: F403
+from .fuse_modules import fuse_modules_qat  # noqa: F403
 from .fuser_method_mappings import *  # noqa: F403
 from .observer import *  # noqa: F403
 from .qconfig import *  # noqa: F403

torch/ao/quantization/fuse_modules.py

@@ -28,7 +28,7 @@ def _set_module(model, submodule_key, module):
 
     setattr(cur_mod, tokens[-1], module)
 
-def fuse_known_modules(mod_list, additional_fuser_method_mapping=None):
+def fuse_known_modules(mod_list, is_qat, additional_fuser_method_mapping=None):
     r"""Returns a list of modules that fuses the operations specified
      in the input module list.
 
@@ -46,7 +46,7 @@ def fuse_known_modules(mod_list, additional_fuser_method_mapping=None):
     if fuser_method is None:
         raise NotImplementedError("Cannot fuse modules: {}".format(types))
     new_mod : List[Optional[nn.Module]] = [None] * len(mod_list)
-    fused = fuser_method(*mod_list)
+    fused = fuser_method(is_qat, *mod_list)
     # NOTE: forward hooks not processed in the two following for loops will be lost after the fusion
     # Move pre forward hooks of the base module to resulting fused module
     for handle_id, pre_hook_fn in mod_list[0]._forward_pre_hooks.items():
@@ -65,7 +65,7 @@ def fuse_known_modules(mod_list, additional_fuser_method_mapping=None):
 
     return new_mod
 
-def _fuse_modules(model, modules_to_fuse, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):
+def _fuse_modules_helper(model, modules_to_fuse, is_qat, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):
     if fuse_custom_config_dict is None:
         fuse_custom_config_dict = {}
     additional_fuser_method_mapping = fuse_custom_config_dict.get("additional_fuser_method_mapping", {})
@@ -74,12 +74,25 @@ def _fuse_modules(model, modules_to_fuse, fuser_func=fuse_known_modules, fuse_cu
         mod_list.append(_get_module(model, item))
 
     # Fuse list of modules
-    new_mod_list = fuser_func(mod_list, additional_fuser_method_mapping)
+    new_mod_list = fuser_func(mod_list, is_qat, additional_fuser_method_mapping)
 
     # Replace original module list with fused module list
     for i, item in enumerate(modules_to_fuse):
         _set_module(model, item, new_mod_list[i])
 
+def _fuse_modules(model, modules_to_fuse, is_qat, inplace=False, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):
+    if not inplace:
+        model = copy.deepcopy(model)
+
+    if all(isinstance(module_element, str) for module_element in modules_to_fuse):
+        # Handle case of modules_to_fuse being a list
+        _fuse_modules_helper(model, modules_to_fuse, is_qat, fuser_func, fuse_custom_config_dict)
+    else:
+        # Handle case of modules_to_fuse being a list of lists
+        for module_list in modules_to_fuse:
+            _fuse_modules_helper(model, module_list, is_qat, fuser_func, fuse_custom_config_dict)
+    return model
+
 def fuse_modules(model, modules_to_fuse, inplace=False, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):
     r"""Fuses a list of modules into a single module
 
@@ -121,27 +134,34 @@ def fuse_modules(model, modules_to_fuse, inplace=False, fuser_func=fuse_known_mo
 
     Examples::
 
-            >>> m = myModel()
+            >>> m = M().eval()
             >>> # m is a module containing the sub-modules below
             >>> modules_to_fuse = [ ['conv1', 'bn1', 'relu1'], ['submodule.conv', 'submodule.relu']]
             >>> fused_m = torch.ao.quantization.fuse_modules(m, modules_to_fuse)
             >>> output = fused_m(input)
 
-            >>> m = myModel()
+            >>> m = M().eval()
             >>> # Alternately provide a single list of modules to fuse
             >>> modules_to_fuse = ['conv1', 'bn1', 'relu1']
             >>> fused_m = torch.ao.quantization.fuse_modules(m, modules_to_fuse)
             >>> output = fused_m(input)
 
     """
-    if not inplace:
-        model = copy.deepcopy(model)
+    return _fuse_modules(
+        model,
+        modules_to_fuse,
+        is_qat=False,
+        inplace=inplace,
+        fuser_func=fuse_known_modules,
+        fuse_custom_config_dict=None)
 
-    if all(isinstance(module_element, str) for module_element in modules_to_fuse):
-        # Handle case of modules_to_fuse being a list
-        _fuse_modules(model, modules_to_fuse, fuser_func, fuse_custom_config_dict)
-    else:
-        # Handle case of modules_to_fuse being a list of lists
-        for module_list in modules_to_fuse:
-            _fuse_modules(model, module_list, fuser_func, fuse_custom_config_dict)
-    return model
+def fuse_modules_qat(model, modules_to_fuse, inplace=False, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):
+    """ QAT version for `fuse_modules`
+    """
+    return _fuse_modules(
+        model,
+        modules_to_fuse,
+        is_qat=True,
+        inplace=inplace,
+        fuser_func=fuse_known_modules,
+        fuse_custom_config_dict=None)

torch/ao/quantization/fuser_method_mappings.py

@@ -7,10 +7,12 @@ from torch.ao.quantization.utils import Pattern
 from torch.ao.quantization.utils import get_combined_dict
 
 
-def fuse_conv_bn(conv, bn):
+def fuse_conv_bn(is_qat, conv, bn):
     r"""Given the conv and bn modules, fuses them and returns the fused module
 
     Args:
+        is_qat: a flag for whether we are using quantization aware training fusion
+        or post training quantization fusion
         conv: Module instance of type conv2d/conv3d
         bn: Spatial BN instance that needs to be fused with the conv
 
@@ -29,7 +31,9 @@ def fuse_conv_bn(conv, bn):
         nn.Conv3d: nni.ConvBn3d,
     }
 
-    if conv.training:
+    if is_qat:
+        # TODO: remove the assert later
+        assert conv.training, "qat is only supported when conv.training is True currently"
         assert bn.num_features == conv.out_channels, 'Output channel of Conv2d must match num_features of BatchNorm2d'
         assert bn.affine, 'Only support fusing BatchNorm2d with affine set to True'
         assert bn.track_running_stats, 'Only support fusing BatchNorm2d with tracking_running_stats set to True'
@@ -41,10 +45,12 @@ def fuse_conv_bn(conv, bn):
     else:
         return nn.utils.fuse_conv_bn_eval(conv, bn)
 
-def fuse_conv_bn_relu(conv, bn, relu):
+def fuse_conv_bn_relu(is_qat, conv, bn, relu):
     r"""Given the conv and bn modules, fuses them and returns the fused module
 
     Args:
+        is_qat: a flag for whether we are using quantization aware training fusion
+        or post training quantization fusion
         conv: Module instance of type conv2d/conv3d
         bn: Spatial BN instance that needs to be fused with the conv
 
@@ -58,7 +64,9 @@ def fuse_conv_bn_relu(conv, bn, relu):
     assert(conv.training == bn.training == relu.training),\
         "Conv and BN both must be in the same mode (train or eval)."
     fused_module : Optional[Type[nn.Sequential]] = None
-    if conv.training:
+    if is_qat:
+        # TODO: remove the assert later
+        assert conv.training, "qat is only supported when conv.training is True currently"
         map_to_fused_module_train = {
             nn.Conv1d: nni.ConvBnReLU1d,
             nn.Conv2d: nni.ConvBnReLU2d,
@@ -85,10 +93,12 @@ def fuse_conv_bn_relu(conv, bn, relu):
         else:
             raise NotImplementedError("Cannot fuse eval modules: {}".format((conv, bn, relu)))
 
-def fuse_linear_bn(linear, bn):
+def fuse_linear_bn(is_qat, linear, bn):
     r"""Given the linear and bn modules, fuses them and returns the fused module
 
     Args:
+        is_qat: a flag for whether we are using quantization aware training fusion
+        or post training quantization fusion
         linear: Module instance of type Linear
         bn: BatchNorm1d instance that needs to be fused with the linear layer
 
@@ -101,13 +111,14 @@ def fuse_linear_bn(linear, bn):
     assert(linear.training == bn.training),\
         "Linear and BN both must be in the same mode (train or eval)."
 
-    if linear.training:
+    if is_qat:
+        # TODO: remove the assert later
+        assert linear.training, "qat is only supported when linear.training is True currently"
         raise Exception("Fusing Linear+BatchNorm not yet supported in training.")
     else:
         return nn.utils.fusion.fuse_linear_bn_eval(linear, bn)
 
-
-def fuse_convtranspose_bn(convt, bn):
+def fuse_convtranspose_bn(is_qat, convt, bn):
     r"""Given ConvTranspose and bn modules, fuses them and returns the fused module
 
     Args:
@@ -124,11 +135,20 @@ def fuse_convtranspose_bn(convt, bn):
     assert(convt.training == bn.training),\
         "ConvTranspose and BN both must be in the same mode (train or eval)."
 
-    if convt.training:
+    if is_qat:
+        assert convt.training, "qat is only supported when convt.training is True currently"
         raise Exception("Fusing ConvTranspose+BatchNorm not yet supported in training.")
     else:
         return nn.utils.fusion.fuse_conv_bn_eval(convt, bn, transpose=True)
 
+def sequential_wrapper2(sequential):
+    """ Given a sequential class for two modules, return a function that takes
+    is_qat, and then two modules as argument, that ignores the is_qat flag
+    and always returns the sequential that combines the two input modules
+    """
+    def fuser_method(is_qat, m1, m2):
+        return sequential(m1, m2)
+    return fuser_method
 
 DEFAULT_OP_LIST_TO_FUSER_METHOD: Dict[Tuple, Union[nn.Sequential, Callable]] = {
     (nn.Conv1d, nn.BatchNorm1d): fuse_conv_bn,
@@ -137,13 +157,13 @@ DEFAULT_OP_LIST_TO_FUSER_METHOD: Dict[Tuple, Union[nn.Sequential, Callable]] = {
     (nn.Conv2d, nn.BatchNorm2d, nn.ReLU): fuse_conv_bn_relu,
     (nn.Conv3d, nn.BatchNorm3d): fuse_conv_bn,
     (nn.Conv3d, nn.BatchNorm3d, nn.ReLU): fuse_conv_bn_relu,
-    (nn.Conv1d, nn.ReLU): nni.ConvReLU1d,
-    (nn.Conv2d, nn.ReLU): nni.ConvReLU2d,
-    (nn.Conv3d, nn.ReLU): nni.ConvReLU3d,
+    (nn.Conv1d, nn.ReLU): sequential_wrapper2(nni.ConvReLU1d),
+    (nn.Conv2d, nn.ReLU): sequential_wrapper2(nni.ConvReLU2d),
+    (nn.Conv3d, nn.ReLU): sequential_wrapper2(nni.ConvReLU3d),
     (nn.Linear, nn.BatchNorm1d): fuse_linear_bn,
-    (nn.Linear, nn.ReLU): nni.LinearReLU,
-    (nn.BatchNorm2d, nn.ReLU): nni.BNReLU2d,
-    (nn.BatchNorm3d, nn.ReLU): nni.BNReLU3d,
+    (nn.Linear, nn.ReLU): sequential_wrapper2(nni.LinearReLU),
+    (nn.BatchNorm2d, nn.ReLU): sequential_wrapper2(nni.BNReLU2d),
+    (nn.BatchNorm3d, nn.ReLU): sequential_wrapper2(nni.BNReLU3d),
     (nn.ConvTranspose1d, nn.BatchNorm1d): fuse_convtranspose_bn,
     (nn.ConvTranspose2d, nn.BatchNorm2d): fuse_convtranspose_bn,
     (nn.ConvTranspose3d, nn.BatchNorm3d): fuse_convtranspose_bn,
@@ -161,13 +181,25 @@ def get_fuser_method(op_list, additional_fuser_method_mapping=None):
     assert fuser_method is not None, "did not find fuser method for: {} ".format(op_list)
     return fuser_method
 
+def reverse_sequential_wrapper2(sequential):
+    """ Given a sequential class for two modules, return a function that takes
+    is_qat, and then two modules as argument, that ignores the is_qat flag
+    and always returns the sequential that combines the two input modules, with
+    the order of two inputs reversed
+    """
+    def fuser_method(is_qat, m1, m2):
+        return sequential(m2, m1)
+    return fuser_method
+
 def reverse2(f):
-    return lambda x, y: f(y, x)
+    def reversed(is_qat, x, y):
+        return f(is_qat, y, x)
+    return reversed
 
 def reverse3(f):
-    def reversed(x, w):
+    def reversed(is_qat, x, w):
         y, z = w
-        return f(z, y, x)
+        return f(is_qat, z, y, x)
     return reversed
 
 DEFAULT_PATTERN_TO_FUSER_METHOD: Dict[Pattern, Union[nn.Sequential, Callable]] = {
@@ -177,13 +209,13 @@ DEFAULT_PATTERN_TO_FUSER_METHOD: Dict[Pattern, Union[nn.Sequential, Callable]] =
     (nn.ReLU, (nn.BatchNorm2d, nn.Conv2d)): reverse3(fuse_conv_bn_relu),
     (nn.BatchNorm3d, nn.Conv3d): reverse2(fuse_conv_bn),
     (nn.ReLU, (nn.BatchNorm3d, nn.Conv3d)): reverse3(fuse_conv_bn_relu),
-    (nn.ReLU, nn.Conv1d): reverse2(nni.ConvReLU1d),
-    (nn.ReLU, nn.Conv2d): reverse2(nni.ConvReLU2d),
-    (nn.ReLU, nn.Conv3d): reverse2(nni.ConvReLU3d),
+    (nn.ReLU, nn.Conv1d): reverse_sequential_wrapper2(nni.ConvReLU1d),
+    (nn.ReLU, nn.Conv2d): reverse_sequential_wrapper2(nni.ConvReLU2d),
+    (nn.ReLU, nn.Conv3d): reverse_sequential_wrapper2(nni.ConvReLU3d),
     (nn.BatchNorm1d, nn.Linear): reverse2(fuse_linear_bn),
-    (nn.ReLU, nn.Linear): reverse2(nni.LinearReLU),
-    (nn.ReLU, nn.BatchNorm2d): reverse2(nni.BNReLU2d),
-    (nn.ReLU, nn.BatchNorm3d): reverse2(nni.BNReLU3d),
+    (nn.ReLU, nn.Linear): reverse_sequential_wrapper2(nni.LinearReLU),
+    (nn.ReLU, nn.BatchNorm2d): reverse_sequential_wrapper2(nni.BNReLU2d),
+    (nn.ReLU, nn.BatchNorm3d): reverse_sequential_wrapper2(nni.BNReLU3d),
     (nn.BatchNorm1d, nn.ConvTranspose1d): reverse2(fuse_convtranspose_bn),
     (nn.BatchNorm2d, nn.ConvTranspose2d): reverse2(fuse_convtranspose_bn),
     (nn.BatchNorm3d, nn.ConvTranspose3d): reverse2(fuse_convtranspose_bn),

torch/ao/quantization/fx/fuse.py

@@ -28,6 +28,7 @@ class Fuser:
     def fuse(
         self,
         model: GraphModule,
+        is_qat: bool,
         fuse_custom_config_dict: Optional[Dict[str, Any]] = None,
         backend_config_dict: Optional[Dict[str, Any]] = None,
     ) -> GraphModule:
@@ -72,7 +73,7 @@ class Fuser:
                 root_node = get_root_node(matched_node_pattern)  # type: ignore[index]
                 env[node.name] = obj.fuse(
                     self, load_arg, root_node, matched_node_pattern,  # type: ignore[arg-type]
-                    fuse_custom_config_dict, fuser_method_mapping)
+                    fuse_custom_config_dict, fuser_method_mapping, is_qat)
             elif maybe_last_node is None:
                 env[node.name] = self.fused_graph.node_copy(node, load_arg)
             # node matched in patterns and is not root is removed here

torch/ao/quantization/fx/fusion_patterns.py

@@ -28,7 +28,8 @@ class FuseHandler(ABC):
              root_node: Node,
              matched_node_pattern: NodePattern,
              fuse_custom_config_dict: Dict[str, Any],
-             fuser_method_mapping: Optional[Dict[Pattern, Union[torch.nn.Sequential, Callable]]]) -> Node:
+             fuser_method_mapping: Optional[Dict[Pattern, Union[torch.nn.Sequential, Callable]]],
+             is_qat: bool) -> Node:
         pass
 
 @register_fusion_pattern((torch.nn.ReLU, torch.nn.Conv1d))
@@ -69,7 +70,8 @@ class DefaultFuseHandler(FuseHandler):
              root_node: Node,
              matched_node_pattern: NodePattern,
              fuse_custom_config_dict: Dict[str, Any],
-             fuser_method_mapping: Optional[Dict[Pattern, Union[torch.nn.Sequential, Callable]]]) -> Node:
+             fuser_method_mapping: Optional[Dict[Pattern, Union[torch.nn.Sequential, Callable]]],
+             is_qat: bool) -> Node:
         additional_fuser_method_mapping = fuse_custom_config_dict.get("additional_fuser_method_mapping", {})
         assert root_node.op == "call_module", "Expecting module node to be a call_module Node"
         root_module = quantizer.modules[root_node.target]
@@ -113,7 +115,7 @@ class DefaultFuseHandler(FuseHandler):
         fuser_method = get_fuser_method_new(matched_module_types, fuser_method_mapping)
         # TODO: change the signature for fuser_method to take matched module patterns
         # as input
-        fused_module = fuser_method(*matched_modules)
+        fused_module = fuser_method(is_qat, *matched_modules)
         # TODO: maybe add a pass to cleanup bn modules?
         setattr(quantizer.modules[module_parent_name], module_name, fused_module)
         return quantizer.fused_graph.node_copy(root_node, load_arg)

torch/ao/quantization/fx/prepare.py

@@ -1198,6 +1198,7 @@ def insert_observers_for_model(
 
 def run_prepare_fx_on_standalone_modules(
     model: torch.nn.Module,
+    is_qat: bool,
     modules: Dict[str, torch.nn.Module],
     matches: Any,
     prepare_custom_config_dict: Dict[str, Any],
@@ -1228,6 +1229,7 @@ def run_prepare_fx_on_standalone_modules(
             prepare(
                 standalone_module,
                 sm_qconfig_dict,
+                is_qat,
                 sm_prepare_config_dict,
                 backend_config_dict=sm_backend_config_dict)
         preserved_attributes = \
@@ -1264,12 +1266,12 @@ def save_state(
 def prepare(
         model: GraphModule,
         qconfig_dict: Any,
+        is_qat: bool,
         node_name_to_scope: Dict[str, Tuple[str, type]],
         prepare_custom_config_dict: Optional[Dict[str, Any]] = None,
         equalization_qconfig_dict: Optional[Dict[str, Any]] = None,
         backend_config_dict: Optional[Dict[str, Any]] = None,
-        is_standalone_module: bool = False,
-        is_qat: bool = False) -> ObservedGraphModule:
+        is_standalone_module: bool = False) -> ObservedGraphModule:
     """ standalone_module means it a submodule that is not inlined in
     parent module, and will be quantized separately as one unit.
 
@@ -1388,7 +1390,7 @@ def prepare(
         "output_quantized_idxs", [])
 
     run_prepare_fx_on_standalone_modules(
-        model, modules, matches, prepare_custom_config_dict, backend_config_dict)
+        model, is_qat, modules, matches, prepare_custom_config_dict, backend_config_dict)
 
     # record names for the set of observed node, so that in convert step
     # we know whether we need to convert a floating point module to reference

torch/ao/quantization/fx/qconfig_utils.py

@@ -11,9 +11,9 @@ from torch.fx import (
 from torch.fx.graph import (
     Graph,
 )
+from torch.nn.intrinsic import _FusedModule
 
 from ..utils import _parent_name
-from ..fuser_method_mappings import DEFAULT_OP_LIST_TO_FUSER_METHOD
 from ..qconfig_dict_utils import (
     get_object_type_qconfig,
     maybe_adjust_qconfig_for_module_type_or_name,
@@ -56,23 +56,29 @@ def update_qconfig_for_fusion(
 
     for node in model.graph.nodes:
         if node.op == 'call_module' and node.target in modules:
-            module_type = type(modules[str(node.target)])
-            if module_type not in list(DEFAULT_OP_LIST_TO_FUSER_METHOD.values()):
+            maybe_fused_module = modules[str(node.target)]
+            if not isinstance(maybe_fused_module, _FusedModule):
                 continue
 
-            for ops, fuser in DEFAULT_OP_LIST_TO_FUSER_METHOD.items():
-                if module_type == fuser:
-                    fused_qconfig = object_type_dict.get(ops[0], None)
+            ops = list(maybe_fused_module._modules.values())
+            fused_qconfig = object_type_dict.get(type(ops[0]), None)
 
             # Raise an error if the modules in the fused module have
             # different qconfigs specified in the qconfig_dict
-                    for op in ops:
-                        if not qconfig_equals(object_type_dict.get(op, None), fused_qconfig):
-                            raise LookupError("During fusion, we need to specify the same " +
-                                              f"qconfigs for both modules in {module_type}.")
+            # TODO: currently it only works for modules,
+            # need to make this work for torch.nn.functional.relu
+            # TODO: currently it only works for object_type configurations,
+            # ideally it should work for different types of configurations,
+            # maybe we want to redesign this part
+            for op in ops[1:]:
+                if not qconfig_equals(object_type_dict.get(type(op), None), fused_qconfig):
+                    raise LookupError(
+                        "During fusion, we need to specify the same " +
+                        f"qconfigs for all module types in {type(maybe_fused_module)} " +
+                        f"offending type: {type(op)}")
 
             if fused_qconfig is not None:
-                        object_type_dict[module_type] = fused_qconfig
+                object_type_dict[type(maybe_fused_module)] = fused_qconfig
 
     return qconfig_dict
 

torch/ao/quantization/quantize_fx.py

@@ -47,6 +47,7 @@ def _swap_ff_with_fxff(model: torch.nn.Module) -> None:
 
 def _fuse_fx(
     graph_module: GraphModule,
+    is_qat: bool,
     fuse_custom_config_dict: Optional[Dict[str, Any]] = None,
     backend_config_dict: Optional[Dict[str, Any]] = None,
 ) -> GraphModule:
@@ -57,7 +58,8 @@ def _fuse_fx(
     """
     _check_is_graph_module(graph_module)
     fuser = Fuser()
-    return fuser.fuse(graph_module, fuse_custom_config_dict, backend_config_dict)
+    return fuser.fuse(
+        graph_module, is_qat, fuse_custom_config_dict, backend_config_dict)
 
 
 class Scope(object):
@@ -175,11 +177,11 @@ class QuantizationTracer(Tracer):
 def _prepare_fx(
     model: torch.nn.Module,
     qconfig_dict: Any,
+    is_qat: bool,
     prepare_custom_config_dict: Optional[Dict[str, Any]] = None,
     equalization_qconfig_dict: Optional[Dict[str, Any]] = None,
     backend_config_dict: Optional[Dict[str, Any]] = None,
     is_standalone_module: bool = False,
-    is_qat: bool = False,
 ) -> ObservedGraphModule:
     r""" Internal helper function for prepare_fx
     Args:
@@ -235,16 +237,20 @@ forward graph of the parent module,
     graph_module = GraphModule(model, tracer.trace(model))
     for attr_name in preserved_attributes:
         setattr(graph_module, attr_name, getattr(model, attr_name))
-    graph_module = _fuse_fx(graph_module, prepare_custom_config_dict, backend_config_dict)
+    graph_module = _fuse_fx(
+        graph_module,
+        is_qat,
+        prepare_custom_config_dict,
+        backend_config_dict)
     prepared = prepare(
         graph_module,
         qconfig_dict,
+        is_qat,
         tracer.node_name_to_scope,
         prepare_custom_config_dict=prepare_custom_config_dict,
         equalization_qconfig_dict=equalization_qconfig_dict,
         backend_config_dict=backend_config_dict,
         is_standalone_module=is_standalone_module,
-        is_qat=is_qat,
     )
 
     for attr_name in preserved_attributes:
@@ -255,9 +261,9 @@ forward graph of the parent module,
 def _prepare_standalone_module_fx(
     model: torch.nn.Module,
     qconfig_dict: Any,
+    is_qat: bool,
     prepare_custom_config_dict: Optional[Dict[str, Any]] = None,
     backend_config_dict: Optional[Dict[str, Any]] = None,
-    is_qat: bool = False,
 ) -> GraphModule:
     r""" [Internal use only] Prepare a standalone module, so that it can be used when quantizing the
     parent module.
@@ -284,10 +290,10 @@ def _prepare_standalone_module_fx(
     return _prepare_fx(
         model,
         qconfig_dict,
+        is_qat,
         prepare_custom_config_dict,
         backend_config_dict=backend_config_dict,
         is_standalone_module=True,
-        is_qat=is_qat,
     )
 
 
@@ -332,7 +338,7 @@ def fuse_fx(
         )
     for attr_name in preserved_attributes:
         setattr(graph_module, attr_name, getattr(model, attr_name))
-    return _fuse_fx(graph_module, fuse_custom_config_dict)
+    return _fuse_fx(graph_module, False, fuse_custom_config_dict)
 
 
 def prepare_fx(
@@ -509,10 +515,10 @@ def prepare_fx(
     return _prepare_fx(
         model,
         qconfig_dict,
+        False,  # is_qat
         prepare_custom_config_dict,
         equalization_qconfig_dict,
         backend_config_dict,
-        is_qat=False,
     )
 
 
@@ -558,9 +564,9 @@ def prepare_qat_fx(
     return _prepare_fx(
         model,
         qconfig_dict,
+        True,  # is_qat
         prepare_custom_config_dict,
         backend_config_dict=backend_config_dict,
-        is_qat=True,
     )
 
 

torch/testing/_internal/common_quantization.py

@@ -1195,6 +1195,10 @@ class AnnotatedConvBnReLUModel(torch.nn.Module):
         return x
 
     def fuse_model(self):
+        # TODO: remove this check and define two fuse_modules function on this module
+        if self.training:
+            torch.quantization.fuse_modules_qat(self, [['conv', 'bn', 'relu']], inplace=True)
+        else:
             torch.quantization.fuse_modules(self, [['conv', 'bn', 'relu']], inplace=True)
 
 class TwoLayerConvModel(torch.nn.Module):
@@ -1464,7 +1468,11 @@ class InnerModule(torch.nn.Module):
                 if isinstance(named_children[idx + 1][1], torch.nn.ReLU):
                     fusable_layers.append([current_name,
                                            named_children[idx + 1][0]])
-        torch.quantization.fuse_modules(self, fusable_layers, inplace=True)
+        # TODO: remove this check and define two fuse_modules function on this module
+        if self.training:
+            torch.ao.quantization.fuse_modules_qat(self, fusable_layers, inplace=True)
+        else:
+            torch.ao.quantization.fuse_modules(self, fusable_layers, inplace=True)
 
 class FunctionalLinear(torch.nn.Module):
     def __init__(self):
@@ -1955,7 +1963,11 @@ class ResNetBase(torch.nn.Module):
         return out
 
     def fuse_model(self):
-        torch.quantization.fuse_modules(self, [['conv1', 'bn1', 'relu1']], inplace=True)
+        # TODO: remove this check and define two fuse_model function on this module
+        if self.training:
+            torch.ao.quantization.fuse_modules_qat(self, [['conv1', 'bn1', 'relu1']], inplace=True)
+        else:
+            torch.ao.quantization.fuse_modules(self, [['conv1', 'bn1', 'relu1']], inplace=True)
 
 class ModelMultipleOps(torch.nn.Module):
     def __init__(self):