[legacy] move engine to legacy (hpcaitech#4560)

* [legacy] move engine to legacy

* [example] fix seq parallel example

* [example] fix seq parallel example

* [test] test gemini pluging hang

* [test] test gemini pluging hang

* [test] test gemini pluging hang

* [test] test gemini pluging hang

* [test] test gemini pluging hang

* [example] update seq parallel requirements

colossalai/builder/builder.py

@@ -71,7 +71,7 @@ def build_gradient_handler(config, model, optimizer):
         optimizer (:class:`torch.optim.Optimizer`): An optimizer object containing parameters for the gradient handler
 
     Returns:
-        An object of :class:`colossalai.engine.BaseGradientHandler`
+        An object of :class:`colossalai.legacy.engine.BaseGradientHandler`
     """
     config_ = config.copy()
     config_['model'] = model

colossalai/initialize.py

@@ -21,9 +21,9 @@
 from colossalai.context import Config, ConfigException, ParallelMode
 from colossalai.context.moe_context import MOE_CONTEXT
 from colossalai.core import global_context as gpc
-from colossalai.engine import Engine
-from colossalai.engine.gradient_accumulation import accumulate_gradient
-from colossalai.engine.schedule import (
+from colossalai.legacy.engine import Engine
+from colossalai.legacy.engine.gradient_accumulation import accumulate_gradient
+from colossalai.legacy.engine.schedule import (
     InterleavedPipelineSchedule,
     NonPipelineSchedule,
     PipelineSchedule,

colossalai/engine/_base_engine.py → colossalai/legacy/engine/_base_engine.py

@@ -8,11 +8,17 @@
 from torch.nn import Module
 from torch.nn.modules.loss import _Loss
 
-from colossalai.engine.gradient_handler import BaseGradientHandler
-from colossalai.engine.schedule import BaseSchedule, InterleavedPipelineSchedule, NonPipelineSchedule, PipelineSchedule
+from colossalai.legacy.engine.gradient_handler import BaseGradientHandler
+from colossalai.legacy.engine.schedule import (
+    BaseSchedule,
+    InterleavedPipelineSchedule,
+    NonPipelineSchedule,
+    PipelineSchedule,
+)
 from colossalai.logging import get_dist_logger
-from colossalai.zero.legacy.gemini import BaseOpHook, register_ophooks_recursively
 from colossalai.nn.optimizer import ColossalaiOptimizer
+from colossalai.zero.legacy.gemini import BaseOpHook, register_ophooks_recursively
+
 
 class Engine:
     """Basic engine class for training and evaluation. It runs a specific process method

colossalai/engine/gradient_accumulation/__init__.py → colossalai/legacy/engine/gradient_accumulation/__init__.py

@@ -4,7 +4,7 @@
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import _LRScheduler
 
-from colossalai.engine import BaseGradientHandler
+from colossalai.legacy.engine import BaseGradientHandler
 
 from ._gradient_accumulation import (
     GradAccumDataloader,
@@ -33,7 +33,7 @@ def accumulate_gradient(model: nn.Module,
         dataloader (:class:`torch.utils.data.DataLoader` or iterable objects):
             your dataloader object, would be called like iter(dataloader)
         accumulate_size (int): the number of steps to accumulate gradients
-        gradient_handlers (List[:class:`colossalai.engine.BaseGradientHandler`]):
+        gradient_handlers (List[:class:`colossalai.legacy.engine.BaseGradientHandler`]):
             list of gradient handler objects. Default is None.
         lr_scheduler (`torch.optim.lr_scheduler` or `colossalai.nn.lr_scheduler`):
             your ``lr_scheduler`` object for gradient accumulation. Defaults to None.

colossalai/engine/gradient_accumulation/_gradient_accumulation.py → colossalai/legacy/engine/gradient_accumulation/_gradient_accumulation.py

@@ -10,7 +10,7 @@
 from torch.optim.lr_scheduler import _LRScheduler
 from torch.utils.data import DataLoader
 
-from colossalai.engine import BaseGradientHandler
+from colossalai.legacy.engine import BaseGradientHandler
 from colossalai.nn.optimizer import ColossalaiOptimizer
 from colossalai.utils import conditional_context
 
@@ -262,7 +262,7 @@ class GradAccumGradientHandler:
     before accumulation size is reached.
 
     Args:
-        grad_handler (:class:`colossalai.engine.BaseGradientHandler`):
+        grad_handler (:class:`colossalai.legacy.engine.BaseGradientHandler`):
             Your ``gradient_handler`` object for gradient accumulation, would be called when achieving `accumulate_size`.
         accumulate_size (int): The number of steps to accumulate gradients.
 

colossalai/engine/gradient_handler/_data_parallel_gradient_handler.py → colossalai/legacy/engine/gradient_handler/_data_parallel_gradient_handler.py

@@ -1,7 +1,7 @@
+from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
 from colossalai.registry import GRADIENT_HANDLER
 
-from ...context.parallel_mode import ParallelMode
 from ._base_gradient_handler import BaseGradientHandler
 from .utils import bucket_allreduce
 

colossalai/engine/gradient_handler/_moe_gradient_handler.py → colossalai/legacy/engine/gradient_handler/_moe_gradient_handler.py

@@ -1,9 +1,9 @@
 from colossalai.context.moe_context import MOE_CONTEXT
+from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
 from colossalai.registry import GRADIENT_HANDLER
 from colossalai.utils.moe import get_moe_epsize_param_dict
 
-from ...context.parallel_mode import ParallelMode
 from ._base_gradient_handler import BaseGradientHandler
 from .utils import bucket_allreduce
 

colossalai/engine/gradient_handler/_sequence_parallel_gradient_handler.py → colossalai/legacy/engine/gradient_handler/_sequence_parallel_gradient_handler.py

@@ -1,7 +1,7 @@
+from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
 from colossalai.registry import GRADIENT_HANDLER
 
-from ...context.parallel_mode import ParallelMode
 from ._base_gradient_handler import BaseGradientHandler
 from .utils import bucket_allreduce
 

colossalai/engine/schedule/_base_schedule.py → colossalai/legacy/engine/schedule/_base_schedule.py

@@ -95,7 +95,7 @@ def forward_backward_step(self,
         """The process function over a batch of dataset for training or evaluation.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             data_iter (Iterable): Data iterator from which get a batch of data, obtained by calling iter(dataloader).
             forward_only (bool): If True, the process won't include backward.
             return_loss (bool, optional): If False, the loss won't be returned.

colossalai/engine/schedule/_non_pipeline_schedule.py → colossalai/legacy/engine/schedule/_non_pipeline_schedule.py

@@ -54,7 +54,7 @@ def forward_backward_step(self,
         The returned labels and loss will None if :attr:`return_loss` is False.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             data_iter (Iterable): Dataloader as the form of an iterator, obtained by calling iter(dataloader).
             forward_only (bool, optional):
                 If True, the model is run for the forward pass, else back propagation will be executed.

colossalai/engine/schedule/_pipeline_schedule.py → colossalai/legacy/engine/schedule/_pipeline_schedule.py

@@ -236,7 +236,7 @@ def _forward_step(self, engine, input_obj, return_tensors, return_output_label=T
         Returns output tensor. This is a helper function and can be ignored by users.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             input_obj (Union[:class:`torch.Tensor`, List[:class:`torch.Tensor`]]): Input tensor for this pipeline stage.
             return_tensors (List[:class:`torch.Tensor`]): A list of tensors to return.
             return_output_label (bool, optional): Whether returns output labels.
@@ -274,7 +274,7 @@ def _backward_step(self, engine, input_obj, output_obj, output_obj_grad):
         This is a helper function and can be ignored by users.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             input_obj (Union[:class:`torch.Tensor`, List[:class:`torch.Tensor`]]): input tensor for this pipeline stage.
             output_obj (Union[:class:`torch.Tensor`, List[:class:`torch.Tensor`]]): output tensor for this pipeline stage.
             output_obj_grad (Union[:class:`torch.Tensor`, List[:class:`torch.Tensor`]]): gradient of output tensor for this pipeline stage.
@@ -314,7 +314,7 @@ def forward_backward_step(self, engine, data_iter, forward_only=False, return_lo
         Returns a tuple with losses if the last stage, an empty tuple otherwise.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             data_iter (Iterable): Dataloader as the form of an iterator, obtained by calling iter(dataloader).
             forward_only (bool, optional):
                 Whether run forward step only. Default is false. If true, no backward will be run.
@@ -518,7 +518,7 @@ def _forward_step(self,
         Returns output tensor. This is a helper function and can be ignored by users.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             model_chunk_id (int): The id of model chunks.
             input_obj (Union[:class:`torch.Tensor`, List[:class:`torch.Tensor`]]): Input tensor for this pipeline stage.
             return_tensors (List[:class:`torch.Tensor`]): A list of tensors to return.
@@ -555,7 +555,7 @@ def forward_backward_step(self, engine, data_iter, forward_only=False, return_lo
         communication between pipeline stages as needed.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             data_iter (Iterable): Dataloader as the form of an iterator, obtained by calling iter(dataloader).
             forward_only (bool, optional):
                 Whether run forward step only. Default is false. If true, no backward will be run.

colossalai/engine/schedule/_pipeline_schedule_v2.py → colossalai/legacy/engine/schedule/_pipeline_schedule_v2.py

@@ -69,7 +69,7 @@ def forward_backward_step(self,
         Returns a tuple with losses if the last stage, an empty tuple otherwise.
 
         Args:
-            engine (colossalai.engine.Engine): Colossalai engine for training and inference.
+            engine (colossalai.legacy.engine.Engine): Colossalai engine for training and inference.
             data_iter (Iterable): Dataloader as the form of an iterator, obtained by calling iter(dataloader).
             forward_only (bool, optional):
                 Whether run forward step only. Default is false. If true, no backward will be run.

colossalai/legacy/trainer/_trainer.py

@@ -4,7 +4,7 @@
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
-from colossalai.engine import Engine
+from colossalai.legacy.engine import Engine
 from colossalai.legacy.trainer.hooks import BaseHook
 from colossalai.logging import DistributedLogger
 from colossalai.utils import MultiTimer, is_dp_rank_0, is_no_pp_or_last_stage, is_tp_rank_0

colossalai/utils/profiler/profiler.py

@@ -1,17 +1,17 @@
-import os
-from typing import List
-from colossalai.engine import Engine
-from torch.profiler import profile as torch_profile
-from torch.profiler.profiler import ProfilerAction
-from typing import Any, Callable, Iterable, Optional
-from torch.autograd import ProfilerActivity
+import gzip
 import json
 import os
 import tempfile
-import gzip
+from typing import Any, Callable, Iterable, List, Optional
+
+from torch.autograd import ProfilerActivity
+from torch.profiler import profile as torch_profile
+from torch.profiler.profiler import ProfilerAction
+
+from colossalai.legacy.engine import Engine
+from colossalai.logging import get_dist_logger
 from colossalai.utils.profiler.extention import ProfilerExtension
 from colossalai.utils.profiler.stateful_tensor_mem_extention import StatefulTensorMemoryProfilerExtention
-from colossalai.logging import get_dist_logger
 
 
 class profile(torch_profile):

colossalai/utils/profiler/stateful_tensor_mem_extention.py

@@ -1,12 +1,14 @@
 import os
 import threading
 import time
-import torch
 from enum import Enum
 from typing import List
-from colossalai.gemini.stateful_tensor import StatefulTensor
+
+import torch
+
 from colossalai.gemini.ophooks import BaseOpHook
-from colossalai.engine import Engine
+from colossalai.gemini.stateful_tensor import StatefulTensor
+from colossalai.legacy.engine import Engine
 from colossalai.utils.profiler.extention import ProfilerExtension
 
 

docs/source/en/advanced_tutorials/add_your_parallel.md

@@ -92,14 +92,14 @@ follow the steps below to create a new distributed initialization.
 
 Gradient handlers are objects which execute the all-reduce operations on parameters' gradients. As different all-reduce
 strategies may be executed for different kinds of parallelism, users can
-inherit `colossalai.engine.gradient_handler.BaseGradientHandler` to implement their strategies. Currently, the library
+inherit `colossalai.legacy.engine.gradient_handler.BaseGradientHandler` to implement their strategies. Currently, the library
 uses the normal data parallel gradient handler which all-reduces the gradients across data parallel ranks. The data
 parallel gradient handler is added to the engine automatically if data parallel is detected. You can add your own
 gradient handler like below:
 
 ```python
 from colossalai.registry import GRADIENT_HANDLER
-from colossalai.engine import BaseGradientHandler
+from colossalai.legacy.engine import BaseGradientHandler
 
 @GRADIENT_HANDLER.register_module
 class YourGradientHandler(BaseGradientHandler):
@@ -121,4 +121,5 @@ gradient_handlers = [
 
 Schedule entails how to execute a forward and backward pass. Currently, Colossal-AI provides pipeline and non-pipeline
 schedules. If you want to modify how the forward and backward passes are executed, you can
-inherit `colossalai.engine.schedule.BaseSchedule` and implement the `forward_back_step` function.
+inherit `colossalai.legacy.engine.schedule.BaseSchedule` and implement the `forward_back_step` function.
+<!-- doc-test-command: echo  -->

docs/source/en/advanced_tutorials/train_gpt_using_hybrid_parallelism.md

@@ -39,7 +39,7 @@ from colossalai.amp import AMP_TYPE
 from colossalai.builder.pipeline import partition_uniform
 from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.layer.wrapper import PipelineSharedModuleWrapper

docs/source/en/advanced_tutorials/train_vit_using_pipeline_parallelism.md

@@ -35,7 +35,7 @@ import colossalai.nn as col_nn
 import torch
 import torch.nn as nn
 from colossalai.builder import build_pipeline_model
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.legacy.trainer import Trainer, hooks

docs/source/en/advanced_tutorials/train_vit_with_hybrid_parallelism.md

@@ -415,7 +415,7 @@ def build_pipeline_vit(num_layers, num_chunks, device=torch.device('cuda'), **kw
 
 #### Import modules
 ```python
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.utils import MultiTimer
 import os

docs/source/en/features/gradient_handler.md

@@ -29,7 +29,7 @@ To implement a customized gradient handler, you need to follow these steps.
 
 ```python
 from colossalai.registry import GRADIENT_HANDLER
-from colossalai.engine.gradient_handler import BaseGradientHandler
+from colossalai.legacy.engine.gradient_handler import BaseGradientHandler
 
 
 @GRADIENT_HANDLER.register_module
@@ -61,3 +61,4 @@ to demonstrate the use of gradient handler. In this example, we used `DataParall
 ```shell
 python -m torch.distributed.launch --nproc_per_node 4 --master_addr localhost --master_port 29500  train_with_engine.py
 ```
+<!-- doc-test-command: echo  -->

docs/source/zh-Hans/advanced_tutorials/add_your_parallel.md

@@ -81,14 +81,14 @@ Colossal-AI 为用户提供了一个全局 context，使他们能够轻松地管
 ## 梯度 Handler
 
 梯度 handler 是对参数的梯度执行 all-reduce 操作的对象。由于不同的 all-reduce 策略或许在不同的并行中被执行，用户可以继承
-`colossalai.engine.gradient_handler.BaseGradientHandler` 来实现其策略。目前，Colossal-AI 使用普通的数据并行梯度 handler 在数据并行的 rank 间 all-reduce 梯度。
+`colossalai.legacy.engine.gradient_handler.BaseGradientHandler` 来实现其策略。目前，Colossal-AI 使用普通的数据并行梯度 handler 在数据并行的 rank 间 all-reduce 梯度。
 如果数据并行被检测到，梯度 handler 会被自动添加进 engine。
 
 你可以添加你自己的梯度 handler，如下所示：
 
 ```python
 from colossalai.registry import GRADIENT_HANDLER
-from colossalai.engine import BaseGradientHandler
+from colossalai.legacy.engine import BaseGradientHandler
 
 @GRADIENT_HANDLER.register_module
 class YourGradientHandler(BaseGradientHandler):
@@ -109,4 +109,5 @@ gradient_handlers = [
 ## Schedule
 
 Schedule 包含了如何执行前向和后向计算。目前， Colossal-AI 提供了流水和非流水的 schedule。
-如果你想修改前向和后向计算的执行方式，你可以继承 `colossalai.engine.schedule.BaseSchedule` 并实现 `forward_back_step` 函数。
+如果你想修改前向和后向计算的执行方式，你可以继承 `colossalai.legacy.engine.schedule.BaseSchedule` 并实现 `forward_back_step` 函数。
+<!-- doc-test-command: echo  -->

docs/source/zh-Hans/advanced_tutorials/train_gpt_using_hybrid_parallelism.md

@@ -39,7 +39,7 @@ from colossalai.amp import AMP_TYPE
 from colossalai.builder.pipeline import partition_uniform
 from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.layer.wrapper import PipelineSharedModuleWrapper

docs/source/zh-Hans/advanced_tutorials/train_vit_using_pipeline_parallelism.md

@@ -33,7 +33,7 @@ import colossalai.nn as col_nn
 import torch
 import torch.nn as nn
 from colossalai.builder import build_pipeline_model
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.legacy.trainer import Trainer, hooks

docs/source/zh-Hans/advanced_tutorials/train_vit_with_hybrid_parallelism.md

@@ -380,7 +380,7 @@ def build_pipeline_vit(num_layers, num_chunks, device=torch.device('cuda'), **kw
 
 #### 导入模块
 ```python
-from colossalai.engine.schedule import (InterleavedPipelineSchedule,
+from colossalai.legacy.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.utils import MultiTimer
 import os

docs/source/zh-Hans/features/gradient_handler.md

@@ -26,7 +26,7 @@
 
 ```python
 from colossalai.registry import GRADIENT_HANDLER
-from colossalai.engine.gradient_handler import BaseGradientHandler
+from colossalai.legacy.engine.gradient_handler import BaseGradientHandler
 
 
 @GRADIENT_HANDLER.register_module
@@ -57,3 +57,4 @@ gradient_handler = [dict(type='MyGradientHandler')]
 ```shell
 python -m torch.distributed.launch --nproc_per_node 4 --master_addr localhost --master_port 29500  train_with_engine.py
 ```
+<!-- doc-test-command: echo  -->