[tensor] customized op returns ColoTensor (#875)

* [tensor] customized op returns ColoTensor

* polish

* polish code

colossalai/tensor/_ops/__init__.py

@@ -1,5 +1,4 @@
-from .init import colo_uniform
 from .linear import colo_linear
-from .element_wise import colo_mean
+from .element_wise import *
 from .layernorm import colo_layernorm
 from .loss import colo_cross_entropy

colossalai/tensor/_ops/element_wise.py

@@ -29,3 +29,22 @@ def register_elementwise_op(op):
 
 register_elementwise_op(torch.nn.functional.gelu)
 register_elementwise_op(torch.nn.functional.relu)
+
+
+@colo_op_impl(torch.sum)
+def sum_op(types, args=(), kwargs=None, pg=None):
+    """
+    Handles ``__torch_function__`` dispatch for the elementwise op such
+    as ``torch.sum`.
+    This method computes on either a normal tensor or a sharded tensor.
+    """
+    if len(args) > 0:
+        input_tensor = args[0]
+    if kwargs is None:
+        kwargs = {}
+    if 'input' in kwargs:
+        input_tensor = kwargs['input']
+    # Validate types
+    if not isinstance(input_tensor, ColoTensor):
+        raise TypeError("input needs to be a ColoTensor")
+    return ColoTensor.init_from_torch_tensor(torch.sum(input_tensor.torch_tensor()))

colossalai/tensor/_ops/init.py

@@ -1,29 +0,0 @@
-import torch
-from colossalai.tensor.op_wrapper import colo_op_impl
-
-
-def validate_param(param, param_name):
-    if param is None:
-        raise ValueError(f"param: {param_name} shouldn't be None!")
-
-
-@colo_op_impl(torch.nn.init.uniform_)
-def colo_uniform(types, args=(), kwargs=None, pg=None):
-    r"""
-    Fills the Tensor in sharded_tensor.local_shards with values drawn from the uniform
-    distribution :math:`\mathcal{U}(a, b)`.
-    Args:
-        sharded_tensor: tensor sharded across devices
-        a: the lower bound of the uniform distribution
-        b: the upper bound of the uniform distribution
-    """
-    validate_param(kwargs, "kwargs")
-    stateful_tensor = kwargs["tensor"]
-    validate_param(stateful_tensor, "stateful_tensor")
-    a = kwargs['a']
-    validate_param(a, "a")
-    b = kwargs['b']
-    validate_param(b, "b")
-
-    torch.nn.init.uniform_(stateful_tensor.torch_tensor(), a=a, b=b)
-    return stateful_tensor

colossalai/tensor/_ops/linear.py

@@ -6,7 +6,8 @@ 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.tensor import TensorSpec, ComputePattern, ParallelAction
+from colossalai.tensor import ComputePattern
+
 
 @colo_op_impl(torch.nn.functional.linear)
 def colo_linear(types, args, kwargs, pg):
@@ -25,6 +26,7 @@ def colo_linear(types, args, kwargs, pg):
         bias = kwargs.get('bias', None)
 
     if isinstance(bias, ColoTensor):
+        assert bias.shard_spec.num_action == 0, f"We currently only support bias is duplicated among processes in the linear operator"
         bias = bias.torch_tensor()
 
     # Add communication logic before and after linear call.
@@ -34,7 +36,7 @@ def colo_linear(types, args, kwargs, pg):
                 input_tensor = input_tensor.torch_tensor()
             if isinstance(weight, ColoTensor):
                 weight = weight.torch_tensor()
-            return torch.nn.functional.linear(input_tensor, weight, bias)
+            return ColoTensor.init_from_torch_tensor(torch.nn.functional.linear(input_tensor, weight, bias))
         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
@@ -54,8 +56,7 @@ def colo_linear(types, args, kwargs, pg):
                 output = reduce_input(partial_output, ParallelMode.PARALLEL_1D)
                 # Bias
                 if bias is not None:
-                    bias_ = bias
+                    output = output + bias
-                    output = output + bias_
                 return ColoTensor.init_from_torch_tensor(output)
             else:
                 raise NotImplementedError

colossalai/tensor/spec.py

@@ -2,6 +2,7 @@ from enum import Enum
 from typing import Tuple, List
 from colossalai.context.parallel_mode import ParallelMode
 
+
 class ComputePattern(Enum):
     TP1DRow = 1
     TP1DCol = 2
@@ -10,6 +11,7 @@ class ComputePattern(Enum):
 
 
 class ParallelAction(object):
+
     def __init__(self, priority=0, compute_pattern=ComputePattern.DP, parallel_mode=ParallelMode.DATA) -> None:
         self.priority = priority
         self.compute_pattern = compute_pattern
@@ -24,6 +26,7 @@ class TensorSpec(object):
     parallel computation pattern of the Operator (Layer).
     We have to consider the hybrid parallel mode.
     """
+
     # a list of parallel actions.
     # For example: On 8 GPUs, a hybrid parallel strategy is applied using
     # using ZeRO with DP-degree = 4 and 1DRowTP with TP-degree = 2.
@@ -38,6 +41,7 @@ class TensorSpec(object):
     # 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.
+
     def __init__(self, parallel_action_list: List[ParallelAction] = []):
         self._parallel_action_list = parallel_action_list
         self.sort()
@@ -56,8 +60,8 @@ class TensorSpec(object):
 
     def sort(self):
         if len(self._parallel_action_list) > 0:
-            self._parallel_action_list.sort(key=lambda parallel_action : parallel_action.priority)
+            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:

tests/test_tensor/test_net_tp.py

@@ -1,19 +1,13 @@
-from cProfile import label
-from statistics import mode
-from colossalai.tensor.colo_tensor import ColoTensor
 from tests.components_to_test.registry import non_distributed_component_funcs
 
 import colossalai
 import pytest
-import torch
 import torch.multiprocessing as mp
 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.utils import ColoInitContext
 
-import torch.distributed as dist
 from functools import partial
 
 

tests/test_tensor/test_op.py

@@ -53,11 +53,11 @@ def test_linear():
 
     # torch.nn.functional.linear(torch.randn(1, in_dim), sharded_weight, sharded_bias)
     out = fc(input_tensor)
-    loss = out.sum()
+    loss = torch.sum(out)
     loss.backward()
 
     out_ref = fc_ref(input_ref)
-    loss_ref = out_ref.sum()
+    loss_ref = torch.sum(out_ref)
     loss_ref.backward()
 
     assert (loss_ref == loss)