mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-09-26 04:03:58 +00:00
[misc] update pre-commit and run all files (#4752)
* [misc] update pre-commit * [misc] run pre-commit * [misc] remove useless configuration files * [misc] ignore cuda for clang-format
This commit is contained in:
Hongxin Liu
GitHub
@@ -16,11 +16,10 @@ from ..modeling.vit import (
|
||||
)
|
||||
from .base_policy import ModulePolicyDescription, Policy, SubModuleReplacementDescription
|
||||
|
||||
__all__ = ['ViTPolicy', 'ViTModelPolicy', 'ViTForImageClassificationPolicy', 'ViTForMaskedImageModelingPolicy']
|
||||
__all__ = ["ViTPolicy", "ViTModelPolicy", "ViTForImageClassificationPolicy", "ViTForMaskedImageModelingPolicy"]
|
||||
|
||||
|
||||
class ViTPolicy(Policy):
|
||||
|
||||
def config_sanity_check(self):
|
||||
pass
|
||||
|
||||
@@ -28,8 +27,7 @@ class ViTPolicy(Policy):
|
||||
return self.model
|
||||
|
||||
def module_policy(self) -> Dict[Union[str, nn.Module], ModulePolicyDescription]:
|
||||
|
||||
from transformers.models.vit.modeling_vit import ViTEmbeddings, ViTLayer, ViTModel, ViTOutput, ViTSelfAttention
|
||||
from transformers.models.vit.modeling_vit import ViTEmbeddings, ViTLayer, ViTOutput, ViTSelfAttention
|
||||
|
||||
policy = {}
|
||||
|
||||
@@ -38,77 +36,85 @@ class ViTPolicy(Policy):
|
||||
warnings.warn("Vit dosen't support sequence parallelism now, will ignore the sequence parallelism flag.")
|
||||
|
||||
if self.shard_config.enable_tensor_parallelism:
|
||||
policy[ViTEmbeddings] = ModulePolicyDescription(attribute_replacement={},
|
||||
param_replacement=[],
|
||||
sub_module_replacement=[
|
||||
SubModuleReplacementDescription(
|
||||
suffix="dropout",
|
||||
target_module=DropoutForReplicatedInput,
|
||||
)
|
||||
])
|
||||
policy[ViTEmbeddings] = ModulePolicyDescription(
|
||||
attribute_replacement={},
|
||||
param_replacement=[],
|
||||
sub_module_replacement=[
|
||||
SubModuleReplacementDescription(
|
||||
suffix="dropout",
|
||||
target_module=DropoutForReplicatedInput,
|
||||
)
|
||||
],
|
||||
)
|
||||
|
||||
policy[ViTLayer] = ModulePolicyDescription(attribute_replacement={
|
||||
"attention.attention.num_attention_heads":
|
||||
self.model.config.num_attention_heads // self.shard_config.tensor_parallel_size,
|
||||
"attention.attention.all_head_size":
|
||||
self.model.config.hidden_size // self.shard_config.tensor_parallel_size,
|
||||
},
|
||||
param_replacement=[],
|
||||
sub_module_replacement=[
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.query",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.key",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.value",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.dropout",
|
||||
target_module=col_nn.DropoutForParallelInput,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.output.dense",
|
||||
target_module=col_nn.Linear1D_Row,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.output.dropout",
|
||||
target_module=col_nn.DropoutForReplicatedInput,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="intermediate.dense",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="output.dense",
|
||||
target_module=col_nn.Linear1D_Row,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="output.dropout",
|
||||
target_module=col_nn.DropoutForReplicatedInput,
|
||||
),
|
||||
])
|
||||
policy[ViTLayer] = ModulePolicyDescription(
|
||||
attribute_replacement={
|
||||
"attention.attention.num_attention_heads": self.model.config.num_attention_heads
|
||||
// self.shard_config.tensor_parallel_size,
|
||||
"attention.attention.all_head_size": self.model.config.hidden_size
|
||||
// self.shard_config.tensor_parallel_size,
|
||||
},
|
||||
param_replacement=[],
|
||||
sub_module_replacement=[
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.query",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.key",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.value",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.attention.dropout",
|
||||
target_module=col_nn.DropoutForParallelInput,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.output.dense",
|
||||
target_module=col_nn.Linear1D_Row,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="attention.output.dropout",
|
||||
target_module=col_nn.DropoutForReplicatedInput,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="intermediate.dense",
|
||||
target_module=col_nn.Linear1D_Col,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="output.dense",
|
||||
target_module=col_nn.Linear1D_Row,
|
||||
),
|
||||
SubModuleReplacementDescription(
|
||||
suffix="output.dropout",
|
||||
target_module=col_nn.DropoutForReplicatedInput,
|
||||
),
|
||||
],
|
||||
)
|
||||
|
||||
# use flash attention
|
||||
if self.shard_config.enable_flash_attention:
|
||||
self.append_or_create_method_replacement(description={
|
||||
'forward': get_vit_flash_self_attention_forward(),
|
||||
},
|
||||
policy=policy,
|
||||
target_key=ViTSelfAttention)
|
||||
self.append_or_create_method_replacement(
|
||||
description={
|
||||
"forward": get_vit_flash_self_attention_forward(),
|
||||
},
|
||||
policy=policy,
|
||||
target_key=ViTSelfAttention,
|
||||
)
|
||||
|
||||
# use jit fused operator
|
||||
if self.shard_config.enable_jit_fused:
|
||||
self.append_or_create_method_replacement(description={
|
||||
'forward': get_jit_fused_vit_output_forward(),
|
||||
'dropout_add': get_jit_fused_dropout_add_func(),
|
||||
},
|
||||
policy=policy,
|
||||
target_key=ViTOutput)
|
||||
self.append_or_create_method_replacement(
|
||||
description={
|
||||
"forward": get_jit_fused_vit_output_forward(),
|
||||
"dropout_add": get_jit_fused_dropout_add_func(),
|
||||
},
|
||||
policy=policy,
|
||||
target_key=ViTOutput,
|
||||
)
|
||||
return policy
|
||||
|
||||
def new_model_class(self):
|
||||
@@ -121,7 +127,7 @@ class ViTPolicy(Policy):
|
||||
"""Get pipeline layers for current stage."""
|
||||
assert self.pipeline_stage_manager is not None, "pipeline_stage_manager is None"
|
||||
|
||||
if self.model.__class__.__name__ == 'ViTModel':
|
||||
if self.model.__class__.__name__ == "ViTModel":
|
||||
module = self.model
|
||||
else:
|
||||
module = self.model.vit
|
||||
@@ -138,22 +144,21 @@ class ViTPolicy(Policy):
|
||||
def set_pipeline_forward(self, model_cls: nn.Module, pipeline_forward: Callable, policy: Dict):
|
||||
if self.pipeline_stage_manager:
|
||||
stage_manager = self.pipeline_stage_manager
|
||||
if self.model.__class__.__name__ == 'ViTModel':
|
||||
if self.model.__class__.__name__ == "ViTModel":
|
||||
module = self.model
|
||||
else:
|
||||
module = self.model.vit
|
||||
|
||||
layers_per_stage = Policy.distribute_layers(len(module.encoder.layer), stage_manager.num_stages)
|
||||
stage_index = Policy.get_stage_index(layers_per_stage, stage_manager.stage)
|
||||
method_replacement = {'forward': pipeline_forward(stage_manager=stage_manager, stage_index=stage_index)}
|
||||
self.append_or_create_method_replacement(description=method_replacement,
|
||||
policy=policy,
|
||||
target_key=model_cls)
|
||||
method_replacement = {"forward": pipeline_forward(stage_manager=stage_manager, stage_index=stage_index)}
|
||||
self.append_or_create_method_replacement(
|
||||
description=method_replacement, policy=policy, target_key=model_cls
|
||||
)
|
||||
|
||||
|
||||
# ViTModel
|
||||
class ViTModelPolicy(ViTPolicy):
|
||||
|
||||
def __init__(self) -> None:
|
||||
super().__init__()
|
||||
|
||||
@@ -181,26 +186,29 @@ class ViTModelPolicy(ViTPolicy):
|
||||
|
||||
# ViTForImageClassification
|
||||
class ViTForImageClassificationPolicy(ViTPolicy):
|
||||
|
||||
def module_policy(self):
|
||||
from transformers.models.vit.modeling_vit import ViTForImageClassification, ViTModel
|
||||
|
||||
policy = super().module_policy()
|
||||
if self.shard_config.enable_tensor_parallelism:
|
||||
new_item = {
|
||||
ViTForImageClassification:
|
||||
ModulePolicyDescription(sub_module_replacement=[
|
||||
ViTForImageClassification: ModulePolicyDescription(
|
||||
sub_module_replacement=[
|
||||
SubModuleReplacementDescription(
|
||||
suffix="classifier", target_module=Linear1D_Col, kwargs=dict(gather_output=True))
|
||||
])
|
||||
suffix="classifier", target_module=Linear1D_Col, kwargs=dict(gather_output=True)
|
||||
)
|
||||
]
|
||||
)
|
||||
}
|
||||
policy.update(new_item)
|
||||
|
||||
if self.shard_config.pipeline_stage_manager is not None:
|
||||
self.set_pipeline_forward(model_cls=ViTModel, pipeline_forward=ViTModel_pipeline_forward, policy=policy)
|
||||
self.set_pipeline_forward(model_cls=ViTForImageClassification,
|
||||
pipeline_forward=ViTForImageClassification_pipeline_forward,
|
||||
policy=policy)
|
||||
self.set_pipeline_forward(
|
||||
model_cls=ViTForImageClassification,
|
||||
pipeline_forward=ViTForImageClassification_pipeline_forward,
|
||||
policy=policy,
|
||||
)
|
||||
|
||||
return policy
|
||||
|
||||
@@ -219,7 +227,6 @@ class ViTForImageClassificationPolicy(ViTPolicy):
|
||||
|
||||
# ViTForMaskedImageModeling
|
||||
class ViTForMaskedImageModelingPolicy(ViTPolicy):
|
||||
|
||||
def __init__(self) -> None:
|
||||
super().__init__()
|
||||
|
||||
@@ -230,9 +237,11 @@ class ViTForMaskedImageModelingPolicy(ViTPolicy):
|
||||
|
||||
if self.shard_config.pipeline_stage_manager is not None:
|
||||
self.set_pipeline_forward(model_cls=ViTModel, pipeline_forward=ViTModel_pipeline_forward, policy=policy)
|
||||
self.set_pipeline_forward(model_cls=ViTForMaskedImageModeling,
|
||||
pipeline_forward=ViTForMaskedImageModeling_pipeline_forward,
|
||||
policy=policy)
|
||||
self.set_pipeline_forward(
|
||||
model_cls=ViTForMaskedImageModeling,
|
||||
pipeline_forward=ViTForMaskedImageModeling_pipeline_forward,
|
||||
policy=policy,
|
||||
)
|
||||
return policy
|
||||
|
||||
def get_held_layers(self) -> List[nn.Module]:
|
||||
|
||||
Reference in New Issue
Block a user