[hotfix] Raise messages for indivisible batch sizes with tensor parallelism (#622)

colossalai/nn/layer/colossalai_layer/_utils.py

@@ -1,12 +1,12 @@
 import torch.nn as nn
 from torch import Tensor
 
-from ..parallel_2d._operation import split_tensor_2d
-from ..parallel_2p5d._operation import split_tensor_2p5d
+from ..parallel_2d._operation import split_batch_2d
+from ..parallel_2p5d._operation import split_batch_2p5d
 from ..parallel_3d._operation import split_batch_3d
 from ..utils import get_tensor_parallel_mode
 
-_parallel_split_batch = {'2d': split_tensor_2d, '2.5d': split_tensor_2p5d, '3d': split_batch_3d}
+_parallel_split_batch = {'2d': split_batch_2d, '2.5d': split_batch_2p5d, '3d': split_batch_3d}
 
 
 def partition_batch(input_) -> Tensor:

colossalai/nn/layer/parallel_2d/__init__.py

@@ -1,8 +1,8 @@
-from ._operation import reduce_by_batch_2d, split_tensor_2d
+from ._operation import reduce_by_batch_2d, split_batch_2d
 from .layers import (Classifier2D, Embedding2D, LayerNorm2D, Linear2D, PatchEmbedding2D, VocabParallelClassifier2D,
                      VocabParallelEmbedding2D)
 
 __all__ = [
-    'split_tensor_2d', 'reduce_by_batch_2d', 'Linear2D', 'LayerNorm2D', 'Classifier2D', 'PatchEmbedding2D',
+    'split_batch_2d', 'reduce_by_batch_2d', 'Linear2D', 'LayerNorm2D', 'Classifier2D', 'PatchEmbedding2D',
     'Embedding2D', 'VocabParallelEmbedding2D', 'VocabParallelClassifier2D'
 ]

colossalai/nn/layer/parallel_2d/_operation.py

@@ -720,7 +720,7 @@ def all_gather_tensor_2d(tensor: Tensor, dim: int, parallel_mode: ParallelMode)
     return _AllGatherTensor2D.apply(tensor, dim, parallel_mode)
 
 
-def split_tensor_2d(input_: Tensor, dim: int = 0) -> Tensor:
+def split_batch_2d(input_: Tensor, dim: int = 0) -> Tensor:
     """Splits 2D tensor in specified dimension across cols.
 
     Args:
@@ -730,6 +730,11 @@ def split_tensor_2d(input_: Tensor, dim: int = 0) -> Tensor:
     Returns:
         :class:`torch.tensor`: The tensor has been split.
     """
+    dim_size = input_.size(dim)
+    world_size = gpc.get_world_size(ParallelMode.PARALLEL_2D_COL)
+    assert dim_size % world_size == 0, \
+        f'The batch size ({dim_size}) is not a multiple of 2D size ({world_size}).'
+
     if input_.size(dim) <= 1:
         return input_
     return torch.chunk(input_, gpc.get_world_size(ParallelMode.PARALLEL_2D_COL),
@@ -784,6 +789,11 @@ def reduce_scatter_tensor_2d(tensor: Tensor, dim: int, parallel_mode: ParallelMo
         The parallel_mode should be concluded in ``ParallelMode``. More details about ``ParallelMode`` could be found
         in `parallel_mode <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/context/parallel_mode.py>`_
     """
+    dim_size = tensor.size(dim)
+    world_size = gpc.get_world_size(parallel_mode)
+    assert dim_size % world_size == 0, \
+        f'The batch size ({dim_size}) is not a multiple of 2D size ({world_size}).'
+    
     return _ReduceScatterTensor2D.apply(tensor, dim, parallel_mode)
 
 

colossalai/nn/layer/parallel_2d/layers.py

@@ -19,7 +19,7 @@ from torch.nn import Parameter
 from ..base_layer import ParallelLayer
 from ..utils import divide, set_tensor_parallel_attribute_by_partition, to_2tuple
 from ._operation import (Matmul_AB_2D, Matmul_ABT_2D, add_bias_2d, all_gather_tensor_2d, classifier_2d, layernorm_2d,
-                         reduce_scatter_tensor_2d, split_tensor_2d)
+                         reduce_scatter_tensor_2d, split_batch_2d)
 from ._utils import assert_summa_initialization, get_summa_dim_from_env
 
 
@@ -547,7 +547,7 @@ class PatchEmbedding2D(ParallelLayer):
             destination.update(local_state)
 
     def forward(self, input_: Tensor) -> Tensor:
-        input_ = split_tensor_2d(input_)
+        input_ = split_batch_2d(input_)
 
         B, C, H, W = input_.shape
         assert H == self.img_size[0] and W == self.img_size[1], \
@@ -692,7 +692,7 @@ class Embedding2D(ParallelLayer):
             destination.update(local_state)
 
     def forward(self, input_: Tensor) -> Tensor:
-        input_ = split_tensor_2d(input_)
+        input_ = split_batch_2d(input_)
 
         weight = all_gather_tensor_2d(self.weight, -1, ParallelMode.PARALLEL_2D_COL)
         output = F.embedding(input_, weight, self.padding_idx, *self.embed_args, **self.embed_kwargs)

colossalai/nn/layer/parallel_2p5d/__init__.py

@@ -1,8 +1,8 @@
-from ._operation import reduce_by_batch_2p5d, split_tensor_2p5d
+from ._operation import reduce_by_batch_2p5d, split_batch_2p5d
 from .layers import (Classifier2p5D, Embedding2p5D, LayerNorm2p5D, Linear2p5D, PatchEmbedding2p5D,
                      VocabParallelClassifier2p5D, VocabParallelEmbedding2p5D)
 
 __all__ = [
-    'split_tensor_2p5d', 'reduce_by_batch_2p5d', 'Linear2p5D', 'LayerNorm2p5D', 'Classifier2p5D', 'PatchEmbedding2p5D',
+    'split_batch_2p5d', 'reduce_by_batch_2p5d', 'Linear2p5D', 'LayerNorm2p5D', 'Classifier2p5D', 'PatchEmbedding2p5D',
     'Embedding2p5D', 'VocabParallelClassifier2p5D', 'VocabParallelEmbedding2p5D'
 ]

colossalai/nn/layer/parallel_2p5d/_operation.py

@@ -755,7 +755,7 @@ class SplitFirst(torch.autograd.Function):
         return grad, None, None
 
 
-def split_tensor_2p5d(input_: Tensor, dim: int = 0) -> Tensor:
+def split_batch_2p5d(input_: Tensor, dim: int = 0) -> Tensor:
     """Splits 2P5D tensor in specified dimension across cols.
 
     Args:
@@ -765,6 +765,11 @@ def split_tensor_2p5d(input_: Tensor, dim: int = 0) -> Tensor:
     Returns:
         :class:`torch.tensor`: The tensor has been split.
     """
+    dim_size = input_.size(dim)
+    world_size = gpc.get_world_size(ParallelMode.PARALLEL_2P5D_COL)
+    assert dim_size % world_size == 0, \
+        f'The batch size ({dim_size}) is not a multiple of 2.5D size * depth ({world_size}).'
+
     if input_.size(dim) <= 1:
         return input_
     return torch.chunk(input_, gpc.get_world_size(ParallelMode.PARALLEL_2P5D_COL),
@@ -819,6 +824,11 @@ def reduce_scatter_tensor_2p5d(input_: Tensor, dim: int, parallel_mode: Parallel
         The parallel_mode should be concluded in ``ParallelMode``. More details about ``ParallelMode`` could be found
         in `parallel_mode <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/context/parallel_mode.py>`_
     """
+    dim_size = input_.size(dim)
+    world_size = gpc.get_world_size(parallel_mode)
+    assert dim_size % world_size == 0, \
+        f'The batch size ({dim_size}) is not a multiple of 2.5D size * depth ({world_size}).'
+
     return _ReduceScatterTensor2p5D.apply(input_, dim, parallel_mode)
 
 

colossalai/nn/layer/parallel_2p5d/layers.py

@@ -20,7 +20,7 @@ from torch.nn import Parameter
 from ..base_layer import ParallelLayer
 from ..utils import divide, set_tensor_parallel_attribute_by_partition, to_2tuple
 from ._operation import (Matmul_AB_2p5D, Matmul_ABT_2p5D, add_bias_2p5d, all_gather_tensor_2p5d, classifier_2p5d,
-                         layernorm_2p5d, reduce_scatter_tensor_2p5d, split_tensor_2p5d)
+                         layernorm_2p5d, reduce_scatter_tensor_2p5d, split_batch_2p5d)
 from ._utils import assert_tesseract_initialization, get_tesseract_dim_dep_from_env
 
 
@@ -568,7 +568,7 @@ class PatchEmbedding2p5D(ParallelLayer):
             destination.update(local_state)
 
     def forward(self, input_: Tensor) -> Tensor:
-        input_ = split_tensor_2p5d(input_, 0)
+        input_ = split_batch_2p5d(input_, 0)
 
         B, C, H, W = input_.shape
         assert H == self.img_size[0] and W == self.img_size[1], \
@@ -713,7 +713,7 @@ class Embedding2p5D(ParallelLayer):
             destination.update(local_state)
 
     def forward(self, input_: Tensor) -> Tensor:
-        input_ = split_tensor_2p5d(input_, 0)
+        input_ = split_batch_2p5d(input_, 0)
 
         weight = all_gather_tensor_2p5d(self.weight, -1, ParallelMode.PARALLEL_2P5D_COL)
 

colossalai/nn/layer/parallel_3d/_operation.py

@@ -276,6 +276,11 @@ def split_tensor_3d(tensor: Tensor, dim: int, parallel_mode: ParallelMode) -> Te
         The parallel_mode should be concluded in ``ParallelMode``. More details about ``ParallelMode`` could be found
         in `parallel_mode <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/context/parallel_mode.py>`_.
     """
+    dim_size = tensor.size(dim)
+    world_size = gpc.get_world_size(parallel_mode)
+    assert dim_size % world_size == 0, \
+        f'The dimension {dim} to split, size ({dim_size}) is not a multiple of world size ({world_size}), ' \
+        f'cannot split tensor evenly'
     if tensor.size(dim) <= 1:
         return tensor
     output = torch.chunk(tensor, gpc.get_world_size(parallel_mode),
@@ -302,13 +307,20 @@ def split_batch_3d(input_: Tensor,
         The parallel_mode should be concluded in ``ParallelMode``. More details about ``ParallelMode`` could be found
         in `parallel_mode <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/context/parallel_mode.py>`_.
     """
-    if input_.size(dim) <= 1:
-        return input_
+    dim_size = input_.size(dim)
     weight_parallel_mode = get_parallel_mode_from_env(WEIGHT_GROUP_3D)
     input_parallel_mode = get_parallel_mode_from_env(INPUT_GROUP_3D)
-    output = torch.chunk(input_, gpc.get_world_size(weight_parallel_mode),
+    weight_world_size = gpc.get_world_size(weight_parallel_mode)
+    input_world_size = gpc.get_world_size(input_parallel_mode)
+
+    assert dim_size % (input_world_size*weight_world_size) == 0, \
+        f'The batch size ({dim_size}) is not a multiple of square of 3D depth ({input_world_size*weight_world_size}).'
+
+    if input_.size(dim) <= 1:
+        return input_
+    output = torch.chunk(input_, weight_world_size,
                          dim=dim)[gpc.get_local_rank(weight_parallel_mode)].contiguous()
-    output = torch.chunk(output, gpc.get_world_size(input_parallel_mode),
+    output = torch.chunk(output, input_world_size,
                          dim=dim)[gpc.get_local_rank(input_parallel_mode)].contiguous()
     return output
 
@@ -394,6 +406,11 @@ def reduce_scatter_tensor_3d(tensor: Tensor, dim: int, parallel_mode: ParallelMo
         The parallel_mode should be concluded in ``ParallelMode``. More details about ``ParallelMode`` could be found
         in `parallel_mode <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/context/parallel_mode.py>`_
     """
+    dim_size = tensor.size(dim)
+    world_size = gpc.get_world_size(parallel_mode)
+    assert dim_size % world_size == 0, \
+        f'The batch size ({dim_size}) is not a multiple of square of 3D depth ({world_size}).'
+
     return _ReduceScatterTensor3D.apply(tensor, dim, parallel_mode)
 
 

colossalai/nn/loss/loss_2d.py

@@ -2,7 +2,7 @@ import torch
 import torch.distributed as dist
 from colossalai.context import ParallelMode
 from colossalai.core import global_context as gpc
-from colossalai.nn.layer.parallel_2d import reduce_by_batch_2d, split_tensor_2d
+from colossalai.nn.layer.parallel_2d import reduce_by_batch_2d, split_batch_2d
 from colossalai.nn.layer.parallel_2d._utils import assert_summa_initialization
 from colossalai.registry import LOSSES
 from colossalai.utils import get_current_device
@@ -48,7 +48,7 @@ class CrossEntropyLoss2D(_Loss):
         Returns:
             float: the loss between logits and targets.
         """
-        targets = split_tensor_2d(targets)
+        targets = split_batch_2d(targets)
         loss = cross_entropy(logits, targets, reduction='none', *self.loss_args, **self.loss_kwargs)
         if self.reduction_mean:
             loss = loss.mean()
@@ -145,7 +145,7 @@ class VocabParallelCrossEntropyLoss2D(_Loss):
             logits (:class:`torch.tensor`): Predicted unnormalized scores (often referred to as logits).
             targets (:class:`torch.tensor`): Ground truth class indices or class probabilities.
         """
-        targets = split_tensor_2d(targets)
+        targets = split_batch_2d(targets)
         loss = _VocabParallelCrossEntropy2D.apply(
             logits,
             targets,

colossalai/nn/loss/loss_2p5d.py

@@ -2,7 +2,7 @@ import torch
 import torch.distributed as dist
 from colossalai.context import ParallelMode
 from colossalai.core import global_context as gpc
-from colossalai.nn.layer.parallel_2p5d import reduce_by_batch_2p5d, split_tensor_2p5d
+from colossalai.nn.layer.parallel_2p5d import reduce_by_batch_2p5d, split_batch_2p5d
 from colossalai.nn.layer.parallel_2p5d._utils import assert_tesseract_initialization
 from colossalai.registry import LOSSES
 from colossalai.utils import get_current_device
@@ -44,7 +44,7 @@ class CrossEntropyLoss2p5D(_Loss):
             logits (:class:`torch.tensor`): Predicted unnormalized scores (often referred to as logits).
             targets (:class:`torch.tensor`): Ground truth class indices or class probabilities.
         """
-        targets = split_tensor_2p5d(targets)
+        targets = split_batch_2p5d(targets)
         loss = cross_entropy(logits, targets, reduction='none', *self.loss_args, **self.loss_kwargs)
         if self.reduction_mean:
             loss = loss.mean()
@@ -138,7 +138,7 @@ class VocabParallelCrossEntropyLoss2p5D(_Loss):
             logits (:class:`torch.tensor`): Predicted unnormalized scores (often referred to as logits).
             targets (:class:`torch.tensor`): Ground truth class indices or class probabilities.
         """
-        targets = split_tensor_2p5d(targets)
+        targets = split_batch_2p5d(targets)
         loss = _VocabParallelCrossEntropy2p5D.apply(logits, targets)
         if self.reduction_mean:
             loss = loss.mean()

colossalai/nn/metric/accuracy_2d.py

@@ -1,5 +1,5 @@
 import torch
-from colossalai.nn.layer.parallel_2d import reduce_by_batch_2d, split_tensor_2d
+from colossalai.nn.layer.parallel_2d import reduce_by_batch_2d, split_batch_2d
 from torch import nn
 
 from ._utils import calc_acc
@@ -22,7 +22,7 @@ class Accuracy2D(nn.Module):
             float: the accuracy of prediction.
         """
         with torch.no_grad():
-            targets = split_tensor_2d(targets)
+            targets = split_batch_2d(targets)
             correct = calc_acc(logits, targets)
             correct = reduce_by_batch_2d(correct)
         return correct

colossalai/nn/metric/accuracy_2p5d.py

@@ -1,5 +1,5 @@
 import torch
-from colossalai.nn.layer.parallel_2p5d import reduce_by_batch_2p5d, split_tensor_2p5d
+from colossalai.nn.layer.parallel_2p5d import reduce_by_batch_2p5d, split_batch_2p5d
 from torch import nn
 
 from ._utils import calc_acc
@@ -22,7 +22,7 @@ class Accuracy2p5D(nn.Module):
             float: the accuracy of prediction.
         """
         with torch.no_grad():
-            targets = split_tensor_2p5d(targets)
+            targets = split_batch_2p5d(targets)
             correct = calc_acc(logits, targets)
             correct = reduce_by_batch_2p5d(correct)
         return correct