[Graph] building computing graph with ColoTensor, Linear only (#917)

colossalai/tensor/_ops/linear.py

@@ -1,15 +1,14 @@
 import torch
 from colossalai.tensor.op_wrapper import colo_op_impl
-from colossalai.context import ParallelMode
-from colossalai.nn.layer.parallel_1d._utils import split_forward_gather_backward, reduce_input, \
-    gather_forward_split_backward, reduce_grad
+from colossalai.nn.layer.parallel_1d._utils import split_forward_gather_backward, reduce_input, reduce_grad
 from colossalai.nn.layer.utils import divide
 from colossalai.core import global_context as gpc
 from packaging import version
 from colossalai.tensor import ComputePattern, TensorSpec, ComputePattern, ParallelAction, ColoTensor, ShardPattern
+from colossalai.tensor.graph import GraphOpNode, GraphGlobalEnv
 
 
-def colo_linear_1Drow(input_tensor: ColoTensor, weight: ColoTensor, bias:ColoTensor) -> ColoTensor:
+def colo_linear_1Drow(input_tensor: ColoTensor, weight: ColoTensor, bias: ColoTensor) -> ColoTensor:
     parallel_action = weight.shard_spec.get_action_by_compute_pattern(ComputePattern.TP1DRow_Linear)
     # Input:S[1] x Weight:S[0] = Output:P
     # All-Reduce(Output) + bias = res
@@ -99,20 +98,32 @@ def colo_linear(types, args, kwargs, pg):
     if bias is not None and not isinstance(bias, ColoTensor):
         bias = ColoTensor.init_from_torch_tensor(bias)
 
+    # building the computing graph, inputs -> op
+    if GraphGlobalEnv().graph_building:
+        cur_op_node = GraphOpNode('linear', [weight, bias])
+        cur_op_node.add_prev_tensor(input_tensor)
+
     # Add communication logic before and after linear call.
+    ret_tensor = None
     if not weight.has_spec():    # No Model Parallel Applied
         assert not bias.has_spec(), 'Invalid bias spec for native Linear op'
         input_tensor = input_tensor.torch_tensor()
         weight = weight.torch_tensor()
         bias = bias.torch_tensor()
-        return ColoTensor.init_from_torch_tensor(torch.nn.functional.linear(input_tensor, weight, bias))
+        ret_tensor = ColoTensor.init_from_torch_tensor(torch.nn.functional.linear(input_tensor, weight, bias))
     elif weight.shard_spec.num_action == 1:    # Single Model Parallel Applied
         compute_patterns = weight.shard_spec.compute_patterns
         if ComputePattern.TP1DRow_Linear in compute_patterns:
-            return colo_linear_1Drow(input_tensor, weight, bias)
+            ret_tensor = colo_linear_1Drow(input_tensor, weight, bias)
         elif ComputePattern.TP1DCol_Linear in compute_patterns:
-            return colo_linear_1Dcol(input_tensor, weight, bias)
+            ret_tensor = colo_linear_1Dcol(input_tensor, weight, bias)
         else:
             raise NotImplementedError
     else:
         raise NotImplementedError
+
+    # building the computing graph, op -> output
+    if GraphGlobalEnv().graph_building:
+        cur_op_node.add_post_tensor(ret_tensor)
+
+    return ret_tensor

colossalai/tensor/colo_tensor.py

@@ -38,6 +38,7 @@ class ColoTensor(object):
         self._shard_spec = shard_spec
         self._shard_pattern = ShardPattern.NA
         self._type = TensorType.NONMODEL
+        self._graph_node = None
 
     def __getitem__(self, key):
         return ColoTensor.init_from_torch_tensor(self.torch_tensor()[key])

colossalai/tensor/graph/__init__.py

@@ -0,0 +1,3 @@
+from .graph_node import GraphNode, GraphOpNode, GraphContext, GraphGlobalEnv
+
+__all__ = ['GraphNode', 'GraphOpNode', 'GraphContext', 'GraphGlobalEnv']

colossalai/tensor/graph/graph_node.py

@@ -0,0 +1,97 @@
+from colossalai.tensor import ColoTensor
+from colossalai.context.singleton_meta import SingletonMeta
+
+
+class GraphGlobalEnv(metaclass=SingletonMeta):
+
+    def __init__(self) -> None:
+        self.graph_building = False
+        self.graph_node_list = []
+        self.node_id = -1
+
+    def get_node_id(self):
+        self.node_id += 1
+        return self.node_id
+
+    def add_graph_node(self, node):
+        self.graph_node_list.append(node)
+
+
+class GraphContext():
+    """
+    
+    Building the computing graph under the context
+
+    >>> with GraphContext():
+    >>>     output = model(colo_input_tensor)
+    """
+    graph_nodes = []
+
+    def __enter__(self):
+        GraphGlobalEnv().graph_building = True
+        GraphGlobalEnv().graph_node_list = []
+
+    def __exit__(self, *exc_info):
+        GraphGlobalEnv().graph_building = False
+        GraphGlobalEnv().node_id = -1
+        self.graph_nodes = GraphGlobalEnv().graph_node_list
+
+
+class GraphNode(object):
+
+    def __init__(self) -> None:
+        self.prev_nodes = []
+        self.post_nodes = []
+        self.id = GraphGlobalEnv().get_node_id()
+
+    def add_prev_node(self, node):
+        if GraphGlobalEnv().graph_building:
+            self.prev_nodes.append(node)
+
+    def add_post_node(self, node):
+        if GraphGlobalEnv().graph_building:
+            self.post_nodes.append(node)
+
+    def post_node_empty(self) -> bool:
+        return len(self.post_nodes) == 0
+
+
+class GraphOpNode(GraphNode):
+
+    def __init__(self, op_type, param_list) -> None:
+        super().__init__()
+        self._op_type = op_type
+        self._param_list = param_list
+        GraphGlobalEnv().add_graph_node(self)
+
+    def add_prev_tensor(self, colo_tensor: ColoTensor):
+        r"""
+        Link the current graph op node to previous graph op.
+        Op1 <- Activation (colo_tensor) Op2
+        Op1 <- Op2
+        """
+        if GraphGlobalEnv().graph_building:
+            assert isinstance(colo_tensor, ColoTensor)
+            if colo_tensor._graph_node is None:
+                colo_tensor._graph_node = GraphNode()
+
+            prev_ops = colo_tensor._graph_node.prev_nodes
+            for op_node in prev_ops:
+                self.add_prev_node(op_node)
+                op_node.add_post_node(self)
+
+    def add_post_tensor(self, colo_tensor: ColoTensor):
+        """
+        Op <- Activation (colo_tensor)
+        """
+        if GraphGlobalEnv().graph_building:
+            assert isinstance(colo_tensor, ColoTensor)
+            if colo_tensor._graph_node is None:
+                colo_tensor._graph_node = GraphNode()
+
+            colo_tensor._graph_node.add_prev_node(self)
+
+    def print(self):
+        print(
+            f'GraphOpNode {self._op_type} {self.id}, post nodes {[node.id for node in self.post_nodes]}, prev node number {[node.id for node in self.prev_nodes]}'
+        )

tests/test_tensor/test_graph.py

@@ -0,0 +1,81 @@
+from torch import nn
+import torch
+from colossalai.tensor import ColoTensor
+from colossalai.tensor.graph import GraphContext
+import gc
+
+
+class SimpleNet(nn.Module):
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.proj1 = nn.Linear(4, 8)
+        self.proj2 = nn.Linear(8, 4)
+        self.proj3 = nn.Linear(4, 4)
+        self.proj4 = nn.Linear(4, 4)
+
+    def forward(self, x):
+        x = self.proj1(x)
+        x = self.proj2(x)
+        x = self.proj3(x)
+        x = self.proj4(x)
+        return x
+
+
+def _visit_graph(start_node):
+    if start_node is None:
+        return
+
+    start_node.print()
+
+    post_node_list = start_node.post_nodes
+    for node in post_node_list:
+        _visit_graph(node)
+
+
+def _get_tensors():
+    for obj in gc.get_objects():
+        try:
+            if torch.is_tensor(obj):
+                yield obj
+        except Exception as e:
+            print('A trivial exception occured: {}'.format(e))
+
+
+def _count_tensors():
+    cnt = 0
+    for t in _get_tensors():
+        cnt += 1
+    return cnt
+
+
+def count_tensors(use_colossal):
+    model = SimpleNet()
+
+    model.eval()
+    with torch.no_grad():
+        if use_colossal:
+            colo_input = ColoTensor.init_from_torch_tensor(torch.randn(4))
+            graph_ctx = GraphContext()
+            with graph_ctx:
+                output = model(colo_input)
+            output = model(colo_input)
+            ret = _count_tensors()
+
+            _visit_graph(graph_ctx.graph_nodes[0])
+
+            del graph_ctx
+            return ret
+        else:
+            input_t = torch.randn(4)
+            output = model(input_t)
+            output = model(input_t)
+            return _count_tensors()
+
+
+def test_check_activation_tensors():
+    assert count_tensors(False) == count_tensors(True)
+
+
+if __name__ == "__main__":
+    count_tensors(True)

tests/test_tensor/test_model.py

@@ -26,7 +26,7 @@ from dataclasses import fields
 def _post_init_colo(self):
     class_fields = fields(self)
     # Safety and consistency checks
-    if not len(class_fields):
+    if len(class_fields) == 0:
         raise ValueError(f"{self.__class__.__name__} has no fields.")
     if not all(field.default is None for field in class_fields[1:]):
         raise ValueError(f"{self.__class__.__name__} should not have more than one required field.")
@@ -361,7 +361,7 @@ def run_dist(rank, world_size, port):
     colossalai.launch(config=config, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
     for name in ['simple_net']:
         run_1d_row_tp(name)
-    for name in ['bert', 'simple_net']:    
+    for name in ['bert', 'simple_net']:
         run_1d_hybrid_tp(name)