[Tensor] Add function to spec and update linear 1Drow and unit tests (#869)

colossalai/tensor/__init__.py

@@ -1,7 +1,9 @@
+from .spec import ComputePattern, ParallelAction, TensorSpec
 from .op_wrapper import (
     colo_op_impl,)
 from .colo_tensor import ColoTensor
 from .utils import convert_parameter
 from ._ops import *
 
-__all__ = ['ColoTensor', 'convert_parameter', 'colo_op_impl']
+__all__ = ['ColoTensor', 'convert_parameter', 'colo_op_impl', 'ComputePattern',
+            'TensorSpec', 'ParallelAction']

colossalai/tensor/_ops/__init__.py

@@ -2,4 +2,4 @@ from .init import colo_uniform
 from .linear import colo_linear
 from .element_wise import colo_mean
 from .layernorm import colo_layernorm
-from .loss import colo_cross_entropy
+from .loss import colo_cross_entropy

colossalai/tensor/_ops/linear.py

@@ -6,8 +6,7 @@ from colossalai.nn.layer.parallel_1d._utils import split_forward_gather_backward
 from colossalai.nn.layer.utils import divide
 from colossalai.core import global_context as gpc
 from packaging import version
-from colossalai.utils.cuda import get_current_device
-
+from colossalai.tensor import TensorSpec, ComputePattern, ParallelAction
 
 @colo_op_impl(torch.nn.functional.linear)
 def colo_linear(types, args, kwargs, pg):
@@ -30,32 +29,36 @@ def colo_linear(types, args, kwargs, pg):
 
     # Add communication logic before and after linear call.
     if isinstance(weight, ColoTensor):
-        if weight.shard_spec == None:
+        if weight.shard_spec == None or weight.shard_spec.num_action == 0:
             if isinstance(input_tensor, ColoTensor):
                 input_tensor = input_tensor.torch_tensor()
             if isinstance(weight, ColoTensor):
                 weight = weight.torch_tensor()
             return torch.nn.functional.linear(input_tensor, weight, bias)
-        elif weight.shard_spec == '1Drow':
-            # Input:S[1] x Weight:S[0] = Output:P
-            # All-Reduce(Output) + bias = res
-            assert divide(input_tensor.shape[-1], gpc.tensor_parallel_size) == weight.size(-1), \
-            'Invalid shapes in 1Drow forward: input={}, weight={}. Expected last dim of input {}.'.format(
-                input_tensor.shape, weight.size, weight.size[-1] * gpc.tensor_parallel_size)
-            # Input:S[1]
-            input_per_partition = split_forward_gather_backward(input_tensor, ParallelMode.PARALLEL_1D, dim=-1)
-            # Output:P
-            device = get_current_device()    # TODO where to put to(deivce)?
-            weight_ = weight.torch_tensor().to(device)
-            partial_output = torch.nn.functional.linear(input_per_partition, weight_)
-            # Reduce(Output)
-            output = reduce_input(partial_output, ParallelMode.PARALLEL_1D)
-            # Bias
-            if bias is not None:
-                bias_ = bias.to(device)
-                output = output + bias_
-            return output
-
+        elif weight.shard_spec.num_action == 1:
+            if ComputePattern.TP1DRow in weight.shard_spec.compute_patterns:
+                # Input:S[1] x Weight:S[0] = Output:P
+                # All-Reduce(Output) + bias = res
+                assert divide(input_tensor.shape[-1], gpc.tensor_parallel_size) == weight.size(-1), \
+                'Invalid shapes in 1Drow forward: input={}, weight={}. Expected last dim of input {}.'.format(
+                    input_tensor.shape, weight.size, weight.size(-1) * gpc.tensor_parallel_size)
+                # Input:S[1]
+                if isinstance(input_tensor, ColoTensor):
+                    input_tensor = input_tensor.torch_tensor()
+                parallel_action = weight.shard_spec.get_action_by_compute_pattern(ComputePattern.TP1DRow)
+                input_per_partition = split_forward_gather_backward(input_tensor, parallel_action.parallel_mode, dim=-1)
+                # Output:P
+                weight_ = weight.torch_tensor()
+                partial_output = torch.nn.functional.linear(input_per_partition, weight_)
+                # Reduce(Output)
+                output = reduce_input(partial_output, ParallelMode.PARALLEL_1D)
+                # Bias
+                if bias is not None:
+                    bias_ = bias
+                    output = output + bias_
+                return ColoTensor.init_from_torch_tensor(output)
+            else:
+                raise NotImplementedError
         else:
             raise NotImplementedError
     else:

colossalai/tensor/colo_tensor.py

@@ -1,13 +1,12 @@
+from colossalai.context import parallel_mode
 from .op_wrapper import _COLOSSAL_OPS
 
 import torch
 from typing import Tuple, Optional
 from numpy import product
 from colossalai.core import global_context as gpc
-from colossalai.context import ParallelMode
 from colossalai.nn.layer.utils import divide
-from colossalai.utils.cuda import get_current_device
-
+from colossalai.tensor import TensorSpec, ComputePattern, ParallelAction
 
 class ColoTensor(object):
     """ Data Structure for Tensor in Colossal-AI
@@ -28,7 +27,7 @@ class ColoTensor(object):
             pin_memory=False,
             device=None,
             torch_tensor=torch.empty(0),
-            shard_spec: str = None,
+            shard_spec: TensorSpec = TensorSpec(),
     ):
         self._size = size
         self._dtype = dtype
@@ -39,7 +38,7 @@ class ColoTensor(object):
         self._shard_spec = shard_spec
 
     @property
-    def shard_spec(self) -> Optional[str]:
+    def shard_spec(self) -> TensorSpec:
         return self._shard_spec
 
     @property
@@ -109,27 +108,27 @@ class ColoTensor(object):
                                              device=self._device)
         return self._torch_tensor
 
-    def set_spec(self, spec: str, lazy_shard: bool = False) -> None:
+    def set_spec(self, spec: TensorSpec, lazy_shard: bool = False) -> None:
         self._shard_spec = spec
         if lazy_shard == False:
             self._shard()
 
     def _shard(self):
         assert self._shard_spec is not None, 'You should call set_spec() before _shard() ColoTensor.'
-        if self._shard_spec == "1Drow":    # TODO It actually represents the sharding layout for Linear-1Drow-weight, but we make it simpler now.
-            num_partition = gpc.get_world_size(ParallelMode.TENSOR)
-            local_rank = gpc.get_local_rank(ParallelMode.TENSOR)
-            dim = -1
-            chunk_size = divide(self._size[dim], num_partition)
-            device = get_current_device()
-            # Reshape to get shard for this rank and we don't want autograd
-            # recording here for the narrow op and 'local_shard' should be a
-            # leaf variable in the autograd graph.
-            self._torch_tensor = self._torch_tensor.narrow(dim, local_rank * chunk_size, chunk_size).detach(
-            ).contiguous()    # TODO Shall we clone() here since detach() will point to the old tensor?
-            self._torch_tensor.requires_grad = self._requires_grad
-            self._size = self._torch_tensor.size()
-            self._device = device    # TODO A `fake` device now because torch_tensor.device always = cpu
+        if self._shard_spec.num_action == 1:
+            if ComputePattern.TP1DRow in self._shard_spec.compute_patterns:
+                parallel_action = self._shard_spec.get_action_by_compute_pattern(ComputePattern.TP1DRow)
+                num_partition = gpc.get_world_size(parallel_action.parallel_mode)
+                local_rank = gpc.get_local_rank(parallel_action.parallel_mode)
+                dim = -1
+                chunk_size = divide(self._size[dim], num_partition)
+                # Reshape to get shard for this rank and we don't want autograd
+                # recording here for the narrow op and 'local_shard' should be a
+                # leaf variable in the autograd graph.
+                self._torch_tensor = self._torch_tensor.narrow(dim, local_rank * chunk_size, chunk_size).detach(
+                ).contiguous()    # TODO Shall we clone() here since detach() will point to the old tensor?
+                self._torch_tensor.requires_grad = self._requires_grad
+                self._size = self._torch_tensor.size()
 
     @classmethod
     def __torch_function__(cls, func, types, args=(), kwargs=None):
@@ -151,5 +150,5 @@ class ColoTensor(object):
             kwargs = {k: v.torch_tensor() if isinstance(v, ColoTensor) else v for k, v in kwargs.items()}
             return func(*args, **kwargs)
 
-    def backward(self, retain_graph: bool = False):
-        self._torch_tensor.backward(retain_graph=retain_graph)
+    def backward(self, gradient: Optional[torch.Tensor] = None , retain_graph: bool = False):
+        self._torch_tensor.backward(gradient=gradient, retain_graph=retain_graph)

colossalai/tensor/spec.py

@@ -1,8 +1,6 @@
 from enum import Enum
 from typing import Tuple, List
 from colossalai.context.parallel_mode import ParallelMode
-from colossalai.core import global_context as gpc
-
 
 class ComputePattern(Enum):
     TP1DRow = 1
@@ -12,17 +10,13 @@ class ComputePattern(Enum):
 
 
 class ParallelAction(object):
-    priority = 0
-    compute_pattern = ComputePattern.DP
-    process_group = gpc.get_group(ParallelMode.DATA)
-
-    def __init__(self, priority, compute_pattern, process_group) -> None:
+    def __init__(self, priority=0, compute_pattern=ComputePattern.DP, parallel_mode=ParallelMode.DATA) -> None:
         self.priority = priority
         self.compute_pattern = compute_pattern
-        self.process_group = process_group
+        self.parallel_mode = parallel_mode
 
 
-class TensorSpec(Enum):
+class TensorSpec(object):
     """
     It contains two aspects of information: 
     First, How are tensors distributed in Heterougenous memory space.
@@ -44,4 +38,28 @@ class TensorSpec(Enum):
     # Before Linear Op, we gather the tensors according to ZeRO.
     # We perform Linear Op according to compute pattern of TP1DRow.
     # After Linear Op, we split the tensors according to ZeRO.
-    parallel_action_list: List[ParallelAction] = []
+    def __init__(self, parallel_action_list: List[ParallelAction] = []):
+        self._parallel_action_list = parallel_action_list
+        self.sort()
+
+    @property
+    def parallel_action_list(self):
+        return self._parallel_action_list
+
+    @property
+    def num_action(self):
+        return len(self._parallel_action_list)
+
+    @property
+    def compute_patterns(self):
+        return [parallel_action.compute_pattern for parallel_action in self._parallel_action_list]
+
+    def sort(self):
+        if len(self._parallel_action_list) > 0:
+            self._parallel_action_list.sort(key=lambda parallel_action : parallel_action.priority)
+    
+    def get_action_by_compute_pattern(self, compute_pattern: ComputePattern):
+        for parallel_action in self._parallel_action_list:
+            if parallel_action.compute_pattern == compute_pattern:
+                return parallel_action
+        return None

tests/test_tensor/test_linear_tp.py

@@ -12,6 +12,7 @@ from colossalai.testing import parameterize, rerun_if_address_is_in_use
 from colossalai.utils.cuda import get_current_device
 from colossalai.utils import free_port
 from colossalai.core import global_context as gpc
+from colossalai.tensor import TensorSpec, ComputePattern, ParallelAction
 
 from _utils import check_equal, replace_parameter_add_grad, broadcast_tensor_chunk
 
@@ -45,7 +46,11 @@ def run_linear_tp1d_row_test():
 
     # replace the torch nn.Parameters with ColoTensor
     sharded_weight = ColoTensor.init_from_torch_tensor(W)
-    sharded_weight.set_spec(spec="1Drow") # reshard
+    parallel_action_list = [
+        ParallelAction(priority=1, compute_pattern=ComputePattern.TP1DRow, parallel_mode=ParallelMode.PARALLEL_1D)
+    ]
+    spec = TensorSpec(parallel_action_list)
+    sharded_weight.set_spec(spec=spec) # reshard
     sharded_bias = ColoTensor.init_from_torch_tensor(B)
     replace_parameter_add_grad(layer, sharded_weight, sharded_bias)
     out = layer(A)