[legacy] move trainer to legacy (#4545)

* [legacy] move trainer to legacy

* [doc] update docs related to trainer

* [test] ignore legacy test

colossalai/trainer/_trainer.py → colossalai/legacy/trainer/_trainer.py

-from typing import Union, List, Any
+from typing import Any, List, Union
 
 import torch
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
 from colossalai.engine import Engine
+from colossalai.legacy.trainer.hooks import BaseHook
 from colossalai.logging import DistributedLogger
-from colossalai.utils import MultiTimer
-from colossalai.utils import is_dp_rank_0, is_tp_rank_0, is_no_pp_or_last_stage
-from colossalai.trainer.hooks import BaseHook
+from colossalai.utils import MultiTimer, is_dp_rank_0, is_no_pp_or_last_stage, is_tp_rank_0
 
 
 class Trainer:

colossalai/trainer/hooks/__init__.py → colossalai/legacy/trainer/hooks/__init__.py

 from ._base_hook import BaseHook
 from ._checkpoint_hook import SaveCheckpointHook
-from ._log_hook import (LogMemoryByEpochHook, LogMetricByEpochHook, LogMetricByStepHook, LogTimingByEpochHook,
-                        TensorboardHook)
+from ._log_hook import (
+    LogMemoryByEpochHook,
+    LogMetricByEpochHook,
+    LogMetricByStepHook,
+    LogTimingByEpochHook,
+    TensorboardHook,
+)
 from ._lr_scheduler_hook import LRSchedulerHook
 from ._metric_hook import AccuracyHook, LossHook, MetricHook, ThroughputHook
 

colossalai/trainer/hooks/_checkpoint_hook.py → colossalai/legacy/trainer/hooks/_checkpoint_hook.py

 #!/usr/bin/env python
 # -*- encoding: utf-8 -*-
 import torch
-from colossalai.logging import get_dist_logger
 
+from colossalai.legacy.trainer.hooks import BaseHook
+from colossalai.logging import get_dist_logger
 from colossalai.registry import HOOKS
-from colossalai.trainer.hooks import BaseHook
 from colossalai.utils.checkpointing import save_checkpoint
+
 from ._lr_scheduler_hook import LRSchedulerHook
 
 

colossalai/trainer/hooks/_log_hook.py → colossalai/legacy/trainer/hooks/_log_hook.py

@@ -3,17 +3,17 @@
 
 import os
 import os.path as osp
-
 from typing import List
+
 from colossalai.context import ParallelMode
 from colossalai.core import global_context as gpc
-from colossalai.registry import HOOKS
+from colossalai.legacy.trainer.hooks._metric_hook import ThroughputMetric
 from colossalai.logging import DistributedLogger
-from colossalai.utils import report_memory_usage, is_dp_rank_0, \
-    is_tp_rank_0, is_no_pp_or_last_stage, MultiTimer
+from colossalai.registry import HOOKS
+from colossalai.utils import MultiTimer, is_dp_rank_0, is_no_pp_or_last_stage, is_tp_rank_0, report_memory_usage
+
 from ._base_hook import BaseHook
 from ._commons_ import _format_number
-from colossalai.trainer.hooks._metric_hook import ThroughputMetric
 
 
 class LogByEpochHook(BaseHook):

colossalai/trainer/hooks/_lr_scheduler_hook.py → colossalai/legacy/trainer/hooks/_lr_scheduler_hook.py

-from colossalai.registry import HOOKS
 from torch import Tensor
 
+from colossalai.registry import HOOKS
+
 from ._metric_hook import LearningRateMetric, MetricHook
 
 

colossalai/trainer/hooks/_metric_hook.py → colossalai/legacy/trainer/hooks/_metric_hook.py

@@ -6,6 +6,7 @@ from typing import Callable
 
 import torch
 import torch.distributed as dist
+
 from colossalai.communication import all_reduce
 from colossalai.context import ParallelMode
 from colossalai.core import global_context as gpc
@@ -19,8 +20,8 @@ from ._commons_ import _format_number
 class Metric(ABC):
     """A basic class of metric collectors. It collects a specific
     metric during training or evaluation and would always be used with
-    :class:`MetricHook` to help it update its states and show the 
-    metric. So please use corresponding hook class to make the metric 
+    :class:`MetricHook` to help it update its states and show the
+    metric. So please use corresponding hook class to make the metric
     collector works.
 
     Args:
@@ -220,9 +221,9 @@ class AccuracyMetric(Metric):
 
 
 class MetricHook(BaseHook):
-    """Specialized hook classes for :class:`Metric`. 
-    Some help metric collectors initialize, reset and 
-    update their states. Others are used to display and 
+    """Specialized hook classes for :class:`Metric`.
+    Some help metric collectors initialize, reset and
+    update their states. Others are used to display and
     record the metric.
 
     Args:

docs/source/en/advanced_tutorials/train_gpt_using_hybrid_parallelism.md

@@ -43,7 +43,7 @@ from colossalai.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.layer.wrapper import PipelineSharedModuleWrapper
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.utils.timer import MultiTimer
 from model_zoo.gpt import GPTLMLoss
 from torch.nn import functional as F
@@ -268,3 +268,4 @@ def train():
         return_output_label=False,
     )
 ```
+<!-- doc-test-command: echo  -->

docs/source/en/advanced_tutorials/train_vit_using_pipeline_parallelism.md

@@ -38,7 +38,7 @@ from colossalai.builder import build_pipeline_model
 from colossalai.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer, get_dataloader
 from timm.models import vision_transformer as vit
 from torchvision import transforms
@@ -245,3 +245,4 @@ def train():
                 hooks=hook_list,
                 display_progress=True)
 ```
+<!-- doc-test-command: echo  -->

docs/source/en/advanced_tutorials/train_vit_with_hybrid_parallelism.md

@@ -79,7 +79,7 @@ from colossalai.core import global_context as gpc
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.lr_scheduler import LinearWarmupLR
 from colossalai.nn.metric import Accuracy
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 ```
 
 - Other modules
@@ -644,3 +644,4 @@ torchrun --standalone --nproc_per_node <NUM_GPUs>  train_hybrid.py --config ./co
 # If your torch >= 1.9.0
 # python -m torch.distributed.run --standalone --nproc_per_node= <NUM_GPUs> train_hybrid.py --config ./configs/config_hybrid_parallel.py
 ```
+<!-- doc-test-command: echo  -->

docs/source/en/basics/engine_trainer.md

@@ -64,7 +64,7 @@ Trainer is a more high-level wrapper for the user to execute training with fewer
 
 ```python
 from colossalai.logging import get_dist_logger
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 
 # build components and initialize with colossalai.initialize
 ...
@@ -107,7 +107,7 @@ If you want to customize your own hook class, you can inherit `hooks.BaseHook` a
 
 ```python
 from colossalai.logging import get_dist_logger
-from colossalai.trainer import hooks
+from colossalai.legacy.trainer import hooks
 
 class LogMessageHook(hooks.BaseHook):
 
@@ -345,7 +345,7 @@ If you wish to train with a trainer object, you can follow the code snippet belo
 
 ```python
 from colossalai.nn.metric import Accuracy
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 
 
 # create a trainer object
@@ -387,3 +387,4 @@ python -m torch.distributed.launch --nproc_per_node <num_gpus> --master_addr loc
 # with trainer
 python -m torch.distributed.launch --nproc_per_node <num_gpus> --master_addr localhost --master_port 29500 run_resnet_cifar10_with_trainer.py
 ```
+<!-- doc-test-command: echo  -->

docs/source/en/basics/model_checkpoint.md

@@ -41,7 +41,7 @@ for epoch in range(num_epochs):
 
 #### Save when using trainer
 ```python
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 model = ...
 engine, _, _, _ = colossalai.initialize(model=model, ...)
 trainer = Trainer(engine, ...)
@@ -61,3 +61,4 @@ model = ...
 load_checkpoint('xxx.pt', model)
 ... # train or test
 ```
+<!-- doc-test-command: echo  -->

docs/source/en/features/mixed_precision_training.md

@@ -267,7 +267,7 @@ from pathlib import Path
 from colossalai.core import global_context as gpc
 from colossalai.logging import get_dist_logger
 from colossalai.utils import get_dataloader
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.nn.lr_scheduler import LinearWarmupLR
 from timm.models import vit_base_patch16_224
 from torchvision import datasets, transforms

docs/source/en/features/pipeline_parallel.md

@@ -79,7 +79,7 @@ import colossalai.nn as col_nn
 
 from colossalai.core import global_context as gpc
 from colossalai.logging import disable_existing_loggers, get_dist_logger
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer, get_dataloader
 from colossalai.context import ParallelMode
 from colossalai.pipeline.pipelinable import PipelinableContext
@@ -157,3 +157,4 @@ trainer.fit(train_dataloader=train_dataloader,
 ```
 
 We use `2` pipeline stages and the batch will be split into `4` micro batches.
+<!-- doc-test-command: echo  -->

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

@@ -43,7 +43,7 @@ from colossalai.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.layer.wrapper import PipelineSharedModuleWrapper
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.utils.timer import MultiTimer
 from model_zoo.gpt import GPTLMLoss
 from torch.nn import functional as F
@@ -273,3 +273,4 @@ def train():
         return_output_label=False,
     )
 ```
+<!-- doc-test-command: echo  -->

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

@@ -36,7 +36,7 @@ from colossalai.builder import build_pipeline_model
 from colossalai.engine.schedule import (InterleavedPipelineSchedule,
                                         PipelineSchedule)
 from colossalai.logging import disable_existing_loggers, get_dist_logger
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer, get_dataloader
 from timm.models import vision_transformer as vit
 from torchvision import transforms
@@ -244,3 +244,4 @@ def train():
                 hooks=hook_list,
                 display_progress=True)
 ```
+<!-- doc-test-command: echo  -->

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

@@ -74,7 +74,7 @@ from colossalai.core import global_context as gpc
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.lr_scheduler import LinearWarmupLR
 from colossalai.nn.metric import Accuracy
-from colossalai.trainer import Trainer, hooks
+from colossalai.legacy.trainer import Trainer, hooks
 ```
 
 - 其他模块
@@ -589,3 +589,4 @@ torchrun --standalone --nproc_per_node <NUM_GPUs>  train_hybrid.py --config ./co
 # If your torch >= 1.9.0
 # python -m torch.distributed.run --standalone --nproc_per_node= <NUM_GPUs> train_hybrid.py --config ./configs/config_hybrid_parallel.py
 ```
+<!-- doc-test-command: echo  -->