[autoparallel] add reshape handler v2 and fix some previous bug (#1683)

colossalai/auto_parallel/solver/op_handler/reshape_handler_v2.py

@@ -0,0 +1,35 @@
+import torch
+from .node_handler import NodeHandler
+from ..sharding_strategy import ShardingStrategy_V2, OperationDataType, OperationData, StrategiesVector
+from ..strategy import ReshapeGenerator, StrategyGenerator_V2
+from typing import List, Dict
+from .registry import operator_registry
+import operator
+
+__all__ = ['ReshapeHandler']
+
+
+@operator_registry.register(torch.reshape)
+@operator_registry.register(torch.Tensor.permute)
+class ReshapeHandler(NodeHandler):
+    """
+    A ReshapeHandler which deals with the sharding strategies for Reshape Op, such as torch.reshape.
+    """
+
+    def get_strategy_generator(self) -> List[StrategyGenerator_V2]:
+        op_data_mapping = self.get_operation_data_mapping()
+        generators = []
+        generators.append(ReshapeGenerator(op_data_mapping, self.device_mesh, self.node.args[0]))
+        return generators
+
+    def get_operation_data_mapping(self) -> Dict[str, OperationData]:
+        # use transposed shape for strategies
+        # the strategies will be transformed back to its original shape in self.post_process
+        physical_input_operand = OperationData(name=str(self.node.args[0]),
+                                               type=OperationDataType.ARG,
+                                               data=self.node.args[0]._meta_data)
+        physical_output = OperationData(name=str(self.node), type=OperationDataType.OUTPUT, data=self.node._meta_data)
+
+        mapping = {"input": physical_input_operand, "output": physical_output}
+
+        return mapping

colossalai/auto_parallel/solver/strategy/__init__.py

@@ -5,10 +5,11 @@ from .batch_norm_generator import BatchNormStrategyGenerator
 from .unary_elementwise_generator import UnaryElementwiseGenerator
 from .getitem_generator import GetItemStrategyGenerator, TensorStrategyGenerator, TensorTupleStrategyGenerator
 from .layer_norm_generator import LayerNormGenerator
+from .reshape_generator import ReshapeGenerator
 
 __all__ = [
     'StrategyGenerator_V2', 'DotProductStrategyGenerator', 'MatVecStrategyGenerator',
-    'LinearProjectionStrategyGenerator', 'BatchedMatMulStrategyGenerator', 'ConvStrategyGenerator', 'UnaryElementwiseGenerator',
+    'LinearProjectionStrategyGenerator', 'BatchedMatMulStrategyGenerator', 'ConvStrategyGenerator',
-    'BatchNormStrategyGenerator', 'GetItemStrategyGenerator', 'TensorStrategyGenerator', 'TensorTupleStrategyGenerator',
+    'UnaryElementwiseGenerator', 'BatchNormStrategyGenerator', 'GetItemStrategyGenerator', 'TensorStrategyGenerator',
-    'LayerNormGenerator'
+    'TensorTupleStrategyGenerator', 'LayerNormGenerator', 'ReshapeGenerator'
 ]

colossalai/auto_parallel/solver/strategy/batch_norm_generator.py

@@ -37,7 +37,7 @@ class BatchNormStrategyGenerator(StrategyGenerator_V2):
         assert input_op_data.dim() in (3, 4,
                                        5), f'We suppose the dim of input fed into conv op should in range of [3, 5].'
 
-    def update_compute_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the computation cost per device with this specific strategy.
 
@@ -62,9 +62,9 @@ class BatchNormStrategyGenerator(StrategyGenerator_V2):
             backward_compute_cost += bias_compute_cost
         total_compute_cost = forward_compute_cost + backward_compute_cost
         compute_cost = TrainCycleItem(fwd=forward_compute_cost, bwd=backward_compute_cost, total=total_compute_cost)
-        return compute_cost
+        strategy.compute_cost = compute_cost
 
-    def update_memory_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
         forward_size_mapping = {
             'input': self._compute_size_in_bytes(strategy, "input"),
             'other': self._compute_size_in_bytes(strategy, "other"),

colossalai/auto_parallel/solver/strategy/conv_strategy_generator.py

@@ -29,7 +29,7 @@ class ConvStrategyGenerator(StrategyGenerator_V2):
         assert input_op_data.dim() in (3, 4,
                                        5), f'We suppose the dim of input fed into conv op should in range of [3, 5].'
 
-    def update_compute_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the computation cost per device with this specific strategy.
 
@@ -67,9 +67,9 @@ class ConvStrategyGenerator(StrategyGenerator_V2):
         total_compute_cost = forward_compute_cost + backward_compute_cost
 
         compute_cost = TrainCycleItem(fwd=forward_compute_cost, bwd=backward_compute_cost, total=total_compute_cost)
-        return compute_cost
+        strategy.compute_cost = compute_cost
 
-    def update_memory_cost(self, strategy: ShardingStrategy_V2) -> ShardingStrategy_V2:
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
         forward_size_mapping = {
             'input': self._compute_size_in_bytes(strategy, "input"),
             'other': self._compute_size_in_bytes(strategy, "other"),

colossalai/auto_parallel/solver/strategy/getitem_generator.py

@@ -28,10 +28,11 @@ class GetItemStrategyGenerator(FollowingStrategyGenerator):
     def validate(self) -> bool:
         return super().validate()
 
-    def update_compute_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
-        return TrainCycleItem(fwd=10, bwd=10, total=20)
+        compute_cost = TrainCycleItem(fwd=10, bwd=10, total=20)
+        strategy.compute_cost = compute_cost
 
-    def update_memory_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the memory cost per device with this specific strategy.
         '''
@@ -59,7 +60,6 @@ class GetItemStrategyGenerator(FollowingStrategyGenerator):
                                     parameter=fwd_parameter_cost + bwd_parameter_cost)
         memory_cost = TrainCycleItem(fwd=fwd_mem_cost, bwd=bwd_mem_cost, total=total_mem_cost)
         strategy.memory_cost = memory_cost
-        return super().update_memory_cost(strategy)
 
 
 class TensorStrategyGenerator(GetItemStrategyGenerator):

colossalai/auto_parallel/solver/strategy/layer_norm_generator.py

@@ -23,7 +23,7 @@ class LayerNormGenerator(StrategyGenerator_V2):
     def validate(self) -> bool:
         return super().validate()
 
-    def update_compute_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the computation cost per device with this specific strategy.
 
@@ -52,9 +52,9 @@ class LayerNormGenerator(StrategyGenerator_V2):
             backward_compute_cost += bias_compute_cost
         total_compute_cost = forward_compute_cost + backward_compute_cost
         compute_cost = TrainCycleItem(fwd=forward_compute_cost, bwd=backward_compute_cost, total=total_compute_cost)
-        return compute_cost
+        strategy.compute_cost = compute_cost
 
-    def update_memory_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the memory cost per device with this specific strategy.
         '''
@@ -103,6 +103,9 @@ class LayerNormGenerator(StrategyGenerator_V2):
         total_mesh_dim_list = []
         for mesh_dim_list in dim_partition.values():
             total_mesh_dim_list.extend(mesh_dim_list)
+        # if there is only one sharding dimension, we should use the value instead of list as logical_process_axis.
+        if len(total_mesh_dim_list) == 1:
+            total_mesh_dim_list = total_mesh_dim_list[0]
         communication_action_mapping = {}
 
         other_comm_spec = self.get_communication_spec(

colossalai/auto_parallel/solver/strategy/reshape_generator.py

@@ -0,0 +1,100 @@
+import operator
+from functools import reduce
+from ..sharding_strategy import ShardingStrategy_V2, TrainCycleItem, MemoryCost
+from colossalai.tensor.shape_consistency import CollectiveCommPattern
+from .strategy_generator import FollowingStrategyGenerator
+from typing import List
+import copy
+
+__all__ = ['ReshapeGenerator']
+
+
+class ReshapeGenerator(FollowingStrategyGenerator):
+    """
+    ReshapeGenerator which deals with the sharding strategies of Reshape Op, such as torch.Tensor.permute.
+    """
+
+    def validate(self) -> bool:
+        return super().validate()
+
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
+        compute_cost = TrainCycleItem(fwd=10, bwd=10, total=20)
+        strategy.compute_cost = compute_cost
+
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
+        '''
+        Compute the memory cost per device with this specific strategy.
+        '''
+        forward_size_mapping = {
+            'input': self._compute_size_in_bytes(strategy, "input"),
+            'output': self._compute_size_in_bytes(strategy, "output")
+        }
+
+        backward_size_mapping = copy.deepcopy(forward_size_mapping)
+        backward_size_mapping.pop("output")
+        # compute fwd cost incurred
+        # fwd_cost = input + output
+        fwd_activation_cost = sum([v for k, v in forward_size_mapping.items() if not self.is_param(k)])
+        fwd_parameter_cost = sum([v for k, v in forward_size_mapping.items() if self.is_param(k)])
+        fwd_mem_cost = MemoryCost(activation=fwd_activation_cost, parameter=fwd_parameter_cost)
+
+        # compute bwd cost incurred
+        # bwd_cost = input_grad
+        bwd_activation_cost = sum([v for k, v in backward_size_mapping.items() if not self.is_param(k)])
+        bwd_parameter_cost = sum([v for k, v in backward_size_mapping.items() if self.is_param(k)])
+        bwd_mem_cost = MemoryCost(activation=bwd_activation_cost, parameter=bwd_parameter_cost)
+
+        # compute total cost
+        total_mem_cost = MemoryCost(activation=fwd_activation_cost + bwd_activation_cost,
+                                    parameter=fwd_parameter_cost + bwd_parameter_cost)
+        memory_cost = TrainCycleItem(fwd=fwd_mem_cost, bwd=bwd_mem_cost, total=total_mem_cost)
+        strategy.memory_cost = memory_cost
+
+    def generate(self):
+        strategy_list = []
+        # For reshape function, to keep the computing correctness we keep the sharding
+        # spec of input is fully replicated. In addition, we will keep the output in
+        # replica status and let the successor node choose the way to resharding the
+        # output node. Therefore, the different strategies of input node with same
+        # output sharding spec will generate same strategy for reshape function.
+        for index, strategy in enumerate(self.predecessor_node.strategies_vector):
+            dim_partition_dict_mapping = {}
+            communication_action_mapping = {}
+            input_sharding_spec = strategy.output_sharding_specs[self.op_data["input"]]
+            dim_partition_dict_for_input = input_sharding_spec.dim_partition_dict
+            dim_partition_dict_for_output = {}
+            if isinstance(self.op_data["output"].data, tuple):
+                dim_partition_dict_for_output = [{} for _ in range(len(self.op_data["output"].data))]
+            dim_partition_dict_mapping = {
+                "input": dim_partition_dict_for_input,
+                "output": dim_partition_dict_for_output,
+            }
+            sharding_spec_mapping = self.to_sharding_spec_mapping(dim_partition_dict_mapping)
+            # add index into name to pass the duplicated check
+            # we keep same strategies with different name for node merging, and it will not increase the searching space,
+            # because in solver, this node will be merged into other nodes, and solver will not create a new variable for this node.
+            name = f'{sharding_spec_mapping["input"].sharding_sequence} -> FULLY REPLICATED_{index}'
+
+            total_mesh_dim_list = []
+            for mesh_dim_list in dim_partition_dict_for_input.values():
+                total_mesh_dim_list.extend(mesh_dim_list)
+            # if there is only one sharding dimension, we should use the value instead of list as logical_process_axis.
+            if len(total_mesh_dim_list) == 1:
+                total_mesh_dim_list = total_mesh_dim_list[0]
+
+            input_comm_spec = self.get_communication_spec(
+                sharding_spec=sharding_spec_mapping["input"],
+                communication_pattern=CollectiveCommPattern.GATHER_FWD_SPLIT_BWD,
+                logical_process_axis=total_mesh_dim_list)
+            communication_action_mapping["input"] = input_comm_spec
+            strategy = self.get_sharding_strategy(name=name,
+                                                  sharding_spec_mapping=sharding_spec_mapping,
+                                                  communication_action_mapping=communication_action_mapping)
+            strategy_list.append(strategy)
+
+        for strategy in strategy_list:
+            self.update_communication_cost(strategy)
+            self.update_compute_cost(strategy)
+            self.update_memory_cost(strategy)
+
+        return strategy_list

colossalai/auto_parallel/solver/strategy/strategy_generator.py

@@ -53,9 +53,16 @@ class StrategyGenerator_V2(ABC):
         for op_data_name, dim_partition_dict in mapping.items():
             if op_data_name in self.op_data:
                 op_data = self.op_data[op_data_name]
-                sharding_spec = ShardingSpec(device_mesh=self.device_mesh,
+                if isinstance(op_data.data, tuple) and isinstance(op_data.data[0], torch.Tensor):
-                                             entire_shape=op_data.logical_shape,
+                    sharding_spec = []
-                                             dim_partition_dict=dim_partition_dict)
+                    for output, dim_partition_dict_element in zip(op_data.data, dim_partition_dict):
+                        sharding_spec = ShardingSpec(device_mesh=self.device_mesh,
+                                                     entire_shape=output.shape,
+                                                     dim_partition_dict=dim_partition_dict_element)
+                else:
+                    sharding_spec = ShardingSpec(device_mesh=self.device_mesh,
+                                                 entire_shape=op_data.logical_shape,
+                                                 dim_partition_dict=dim_partition_dict)
                 results[op_data_name] = sharding_spec
         return results
 

colossalai/auto_parallel/solver/strategy/unary_elementwise_generator.py

@@ -18,10 +18,11 @@ class UnaryElementwiseGenerator(FollowingStrategyGenerator):
     def validate(self) -> bool:
         return super().validate()
 
-    def update_compute_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_compute_cost(self, strategy: ShardingStrategy_V2):
-        return TrainCycleItem(fwd=10, bwd=10, total=20)
+        compute_cost = TrainCycleItem(fwd=10, bwd=10, total=20)
+        strategy.compute_cost = compute_cost
 
-    def update_memory_cost(self, strategy: ShardingStrategy_V2) -> TrainCycleItem:
+    def update_memory_cost(self, strategy: ShardingStrategy_V2):
         '''
         Compute the memory cost per device with this specific strategy.
         '''
@@ -49,7 +50,6 @@ class UnaryElementwiseGenerator(FollowingStrategyGenerator):
                                     parameter=fwd_parameter_cost + bwd_parameter_cost)
         memory_cost = TrainCycleItem(fwd=fwd_mem_cost, bwd=bwd_mem_cost, total=total_mem_cost)
         strategy.memory_cost = memory_cost
-        return super().update_memory_cost(strategy)
 
     def generate(self):
         strategy_list = []

tests/test_auto_parallel/test_node_handler/test_reshape_handler_v2.py

@@ -0,0 +1,81 @@
+import torch
+import torch.nn as nn
+from colossalai.fx import ColoTracer, ColoGraphModule
+from colossalai.auto_parallel.solver.op_handler.conv_handler_v2 import ConvFunctionHandler
+from colossalai.auto_parallel.solver.op_handler.reshape_handler_v2 import ReshapeHandler
+from colossalai.auto_parallel.solver.sharding_strategy import OperationData, OperationDataType, StrategiesVector
+from colossalai.device.device_mesh import DeviceMesh
+
+
+class ReshapeModel(nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, input, other):
+        conv_node = nn.functional.conv2d(input, other)
+        reshape_node = conv_node.view(2, -1)
+        return reshape_node
+
+
+def test_reshape_handler():
+    model = ReshapeModel()
+    tracer = ColoTracer()
+    # graph():
+    #     %input_1 : torch.Tensor [#users=1] = placeholder[target=input]
+    #     %other : torch.Tensor [#users=1] = placeholder[target=other]
+    #     %conv2d : [#users=1] = call_function[target=torch.conv2d](args = (%input_1, %other), kwargs = {})
+    #     %view : [#users=1] = call_method[target=view](args = (%conv2d, 2, -1), kwargs = {})
+    #     return view
+    graph = tracer.trace(model,
+                         meta_args={
+                             "input": torch.rand(4, 4, 64, 64).to('meta'),
+                             "other": torch.rand(4, 16, 3, 3).to('meta'),
+                         })
+    gm = ColoGraphModule(model, graph)
+    physical_mesh_id = torch.arange(0, 4)
+
+    mesh_shape = (2, 2)
+    device_mesh = DeviceMesh(physical_mesh_id, mesh_shape)
+    conv_mod_node = list(graph.nodes)[2]
+    reshape_node = list(graph.nodes)[3]
+    reshape_strategies_vector = StrategiesVector(reshape_node)
+    conv_strategies_vector = StrategiesVector(conv_mod_node)
+
+    # build handler
+    conv_handler = ConvFunctionHandler(node=conv_mod_node,
+                                       device_mesh=device_mesh,
+                                       strategies_vector=conv_strategies_vector)
+    conv_handler.register_strategy()
+    setattr(conv_mod_node, 'strategies_vector', conv_strategies_vector)
+    reshape_handler = ReshapeHandler(node=reshape_node,
+                                     device_mesh=device_mesh,
+                                     strategies_vector=reshape_strategies_vector)
+
+    reshape_handler.register_strategy()
+
+    # check operation data mapping
+    mapping = reshape_handler.get_operation_data_mapping()
+
+    for name, op_data in mapping.items():
+        op_data: OperationData
+        # make sure they have valid values
+        assert op_data.data is not None
+
+    assert mapping['input'].name == "conv2d"
+    assert mapping['input'].data.is_meta
+    assert mapping['input'].data.shape == torch.Size([4, 4, 62, 62])
+    assert mapping['input'].type == OperationDataType.ARG
+    assert mapping['input'].logical_shape == torch.Size([4, 4, 62, 62])
+
+    assert mapping['output'].name == "view"
+    assert mapping['output'].data.is_meta
+    assert mapping['output'].data.shape == torch.Size([2, 30752])
+    assert mapping['output'].type == OperationDataType.OUTPUT
+
+    # reshape handler is a following strategy handler, so the number of strategies is equal to the predecessor node.
+    assert len(reshape_strategies_vector) == len(conv_strategies_vector)
+
+
+if __name__ == '__main__':
+    test_reshape_handler()