Loose the group dependency (#4128)

nni/algorithms/compression/pytorch/pruning/dependency_aware_pruner.py

@@ -44,9 +44,8 @@ class DependencyAwarePruner(Pruner):
             self._unwrap_model()
             self.graph = TorchModuleGraph(model, dummy_input)
             self._wrap_model()
-            self.channel_depen = ChannelDependency(
-                traced_model=self.graph.trace)
-            self.group_depen = GroupDependency(traced_model=self.graph.trace)
+            self.channel_depen = ChannelDependency(model, dummy_input, traced_model=self.graph.trace)
+            self.group_depen = GroupDependency(model, dummy_input, traced_model=self.graph.trace)
             self.channel_depen = self.channel_depen.dependency_sets
             self.channel_depen = {
                 name: sets for sets in self.channel_depen for name in sets}

nni/compression/pytorch/utils/__init__.py

-from .utils import *
+from .utils import *
\ No newline at end of file
+from .shape_dependency import *
+from .shape_dependency import ReshapeDependency
+
+def not_safe_to_prune(model, dummy_input):
+    """
+    Get the layers that are not safe to prune(may bring the shape conflict).
+    For example, if the output tensor of a conv layer is directly followed by
+    a shape-dependent function(such as reshape/view), then this conv layer
+    may be not safe to be pruned. Pruning may change the output shape of
+    this conv layer and result in shape problems. This function find all the
+    layers that directly followed by the shape-dependent functions(view, reshape, etc).
+    If you run the inference after the speedup and run into a shape related error,
+    please exclude the layers returned by this function and try again.
+
+    Parameters
+    ----------
+    model: torch.nn.Module
+        The target model to prune.
+    dummy_input: torch.Tensor/list of torch.Tensor/tuple of Tensor
+    """
+    reshape_dset = ReshapeDependency(model, dummy_input)
+    return reshape_dset.dependency_sets

nni/compression/pytorch/utils/mask_conflict.py

@@ -10,7 +10,7 @@ from .utils import get_module_by_name
 _logger = logging.getLogger('FixMaskConflict')
 
 
-def fix_mask_conflict(masks, model=None, dummy_input=None, traced=None):
+def fix_mask_conflict(masks, model, dummy_input, traced=None):
     """
     MaskConflict fix the mask conflict for the channel dependencies
     and group dependency.
@@ -81,7 +81,7 @@ class MaskFix:
 
 
 class GroupMaskConflict(MaskFix):
-    def __init__(self, masks, model=None, dummy_input=None, traced=None):
+    def __init__(self, masks, model, dummy_input, traced=None):
         """
         GroupMaskConflict fix the mask conflict between the layers that
         has group dependecy with each other.
@@ -168,7 +168,7 @@ class GroupMaskConflict(MaskFix):
 
 
 class ChannelMaskConflict(MaskFix):
-    def __init__(self, masks, model=None, dummy_input=None, traced=None):
+    def __init__(self, masks, model, dummy_input, traced=None):
         """
         ChannelMaskConflict fix the mask conflict between the layers that
         has channel dependecy with each other.

nni/compression/pytorch/utils/shape_dependency.py

@@ -3,10 +3,14 @@
 
 import csv
 import logging
+import torch
 import numpy as np
+from nni.compression.pytorch.compressor import PrunerModuleWrapper
+from .utils import get_module_by_name
 
 
-__all__ = ['ChannelDependency', 'GroupDependency', 'InputChannelDependency', 'AttentionWeightDependency']
+__all__ = ['ChannelDependency', 'GroupDependency',
+           'InputChannelDependency', 'AttentionWeightDependency']
 
 
 CONV_TYPE = 'aten::_convolution'
@@ -45,6 +49,7 @@ class Dependency:
             # the model or a already traced model
             assert model is not None and dummy_input is not None
         self.graph = TorchModuleGraph(model, dummy_input, traced_model)
+        self.model = model
         self.dependency = dict()
         self.build_dependency()
 
@@ -85,7 +90,7 @@ def reshape_break_channel_dependency(op_node):
 
 
 class ChannelDependency(Dependency):
-    def __init__(self, model=None, dummy_input=None, traced_model=None, prune_type='Filter'):
+    def __init__(self, model, dummy_input, traced_model=None, prune_type='Filter'):
         """
         This model analyze the channel dependencies between the conv
         layers in a model.
@@ -261,7 +266,7 @@ class InputChannelDependency(ChannelDependency):
     If not, the input channel dependency will be passed to the following nodes.
     """
 
-    def __init__(self, model, dummy_input=None, traced_model=None):
+    def __init__(self, model, dummy_input, traced_model=None):
         """
         This model analyze the input channel dependencies between the conv
         layers in a model.
@@ -323,7 +328,7 @@ class InputChannelDependency(ChannelDependency):
 
 
 class GroupDependency(Dependency):
-    def __init__(self, model=None, dummy_input=None, traced_model=None):
+    def __init__(self, model, dummy_input, traced_model=None):
         """
         This model analyze the group dependencis between the conv
         layers in a model.
@@ -383,13 +388,17 @@ class GroupDependency(Dependency):
         group : int
             the number of the groups of the target conv layer.
         """
-        cpp_conv = list(filter(lambda x: x.kind() ==
-                               CONV_TYPE, node_group.node_cpps))
-        assert len(cpp_conv) == 1
-        cpp_conv = cpp_conv[0]
-        inputs = list(cpp_conv.inputs())
-        # get the number of the group from the input parameters
-        group = inputs[8].toIValue()
+        node_name = node_group.name
+        _, leaf_module = get_module_by_name(self.model, node_name)
+        if isinstance(leaf_module, PrunerModuleWrapper):
+            leaf_module = leaf_module.module
+        assert isinstance(
+            leaf_module, (torch.nn.Conv2d, torch.nn.ConvTranspose2d))
+        group = leaf_module.groups
+        n_filter = leaf_module.out_channels
+        if n_filter == group:
+            # depthwise conv will not introduce extra group dependency
+            return 1
         return group
 
     def build_dependency(self):
@@ -712,4 +721,3 @@ class AttentionWeightDependency(Dependency):
                 group = self.dependency[name]
                 if len(group) > 0:
                     csv_w.writerow([name, group])
-

nni/compression/pytorch/utils/utils.py

 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 import torch
-from .shape_dependency import ReshapeDependency
+
 
 torch_float_dtype = [torch.float, torch.float16, torch.float32, torch.float64, torch.half, torch.double]
 torch_integer_dtype = [torch.uint8, torch.int16, torch.short, torch.int32, torch.long, torch.bool]
@@ -67,23 +67,3 @@ def randomize_tensor(tensor, start=1, end=100):
         # with nn.init.uniform_
         torch.nn.init.uniform_(tensor.data, start, end)
 
-
-def not_safe_to_prune(model, dummy_input):
-    """
-    Get the layers that are not safe to prune(may bring the shape conflict).
-    For example, if the output tensor of a conv layer is directly followed by
-    a shape-dependent function(such as reshape/view), then this conv layer
-    may be not safe to be pruned. Pruning may change the output shape of
-    this conv layer and result in shape problems. This function find all the
-    layers that directly followed by the shape-dependent functions(view, reshape, etc).
-    If you run the inference after the speedup and run into a shape related error,
-    please exclude the layers returned by this function and try again.
-
-    Parameters
-    ----------
-    model: torch.nn.Module
-        The target model to prune.
-    dummy_input: torch.Tensor/list of torch.Tensor/tuple of Tensor
-    """
-    reshape_dset = ReshapeDependency(model, dummy_input)
-    return reshape_dset.dependency_sets
\ No newline at end of file

test/ut/sdk/test_model_speedup.py

@@ -174,10 +174,18 @@ def prune_model_l1(model):
 
 
 def generate_random_sparsity(model):
+    _start = 0.5
+    _end = 0.99
+    if isinstance(model, models.mobilenet.MobileNetV2):
+        # mobilenet models have great propagation characteristics
+        # so we use smaller sparsity ratio to avoid pruning the whole
+        # layer out
+        _start = 0.01
+        _end = 0.3
     cfg_list = []
     for name, module in model.named_modules():
         if isinstance(module, nn.Conv2d):
-            sparsity = np.random.uniform(0.5, 0.99)
+            sparsity = np.random.uniform(_start, _end)
             cfg_list.append({'op_types': ['Conv2d'], 'op_names': [name],
                              'sparsity': sparsity})
     return cfg_list
@@ -187,11 +195,19 @@ def generate_random_sparsity_v2(model):
     """
     Only select 50% layers to prune.
     """
+    _start = 0.5
+    _end = 0.99
+    if isinstance(model, models.mobilenet.MobileNetV2):
+        # mobilenet models have great propagation characteristics
+        # so we use smaller sparsity ratio to avoid pruning the whole
+        # layer out
+        _start = 0.01
+        _end = 0.3
     cfg_list = []
     for name, module in model.named_modules():
         if isinstance(module, nn.Conv2d):
             if np.random.uniform(0, 1.0) > 0.5:
-                sparsity = np.random.uniform(0.5, 0.99)
+                sparsity = np.random.uniform(_start, _end)
                 cfg_list.append({'op_types': ['Conv2d'], 'op_names': [name],
                                  'sparsity': sparsity})
     return cfg_list