* support bf16 in AmpOptimWrapper

* add docstring

* modify docs

* add unittests for bf16 in AmpOptimWrapper

* fix type

* fix to pass ci

* fix ut skip logic to pass ci

* fix as comment

* add type hints

* fix docstring and add warning information

* remove check for pytorch>=1.6 in unittest

* modify unittest

* modify unittest

* remove torch.float32 && torch.float64 from valid dtypes

* fix as comments

* minor refine docstring

* fix unittest parameterized to pass CI

* fix unittest && add back torch.float32, torch.float64

* fix the bug when the params in shared modules do not require grad

* test DefaultOptimWrapperConstructor when the params in shared modules do not require grad

* Fix docstring

* fix

* add ZeroOptimizer to optim

* resolve `duplicate label` warnings

* upgrade docutils && shpinx to resolve `unknown directive or role` warnings

* fix typo

* resolve literal_block && heading warnings

* resolve json literal_block warnings

* resolve python literal_block warnings

* resolve bunches of reference warnings

* resolve bunches of docstring warnings

* resolve warnings in autosummary

* resolve remaining warnings in en docs

* resolve heading warnings in zh_cn docs

* resolve remaining warnings in zh_cn docs

* fix as comments

* fix as comments

* add ApexOptimWrapper

* typo fix

* add apex amp.initialize in optim_context

* assert apex_amp

* polish code

* add parameters of apex_amp.initialize

* add docs

* polish code

* polish code

* polish code

* fix calling of apex amp load_state_dict

* polish

* add comments

* Update apex_optimizer_wrapper.py

* Update apex_optimizer_wrapper.py

---------

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* add dadaptation

* Update mmengine/optim/optimizer/builder.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* update dadaptation docs

---------

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* [Feature] Add ReduceOnPlateauParamScheduler and change ParamSchedulerHook

* [Feature] add ReduceOnPlateauLR and ReduceOnPlateauMomentum

* pre-commit check

* add a little docs

* change position

* fix the conflict between isort and yapf

* fix ParamSchedulerHook after_val_epoch execute without train_loop and param_schedulers built

* Apply suggestions from code review

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>

* update ReduceOnPlateauParamScheduler, ReduceOnPlateauMomentum and ParamSchedulerHook

* fix get need_step_args attribute error in ParamSchedulerHook

* fix load_state_dict error for rule in ReduceOnPlateauParamScheduler

* add docs for ParamSchedulerHook and fix a few codes

* [Docs] add ReduceOnPlateauParamScheduler, ReduceOnPlateauMomentum and ReduceOnPlateauLR docs

* [Refactor] adjust the order of import

* [Fix] add init check for threshold in ReduceOnPlateauParamScheduler

* [Test] add test for ReduceOnPlateauParamScheduler, ReduceOnPlateauLR and ReduceOnPlateauMomentum

* [Fix] fix no attribute self.min_value

* [Fix] fix numerical problem in tests

* [Fix] fix error in tests

* [Fix] fix ignore first param in tests

* [Fix] fix bug in tests

* [Fix] fix bug in tests

* [Fix] fix bug in tests

* [Fix] increase coverage

* [Fix] fix count self._global_step bug and docs

* [Fix] fix tests

* [Fix] modified ParamSchedulerHook test

* Update mmengine/optim/scheduler/param_scheduler.py

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>

* [Fix] modified something according to commented

* [Docs] add api for en and zh_cn

* [Fix] fix bug in test_param_scheduler_hook.py

* [Test] support more complicated test modes(less, greater, rel, abs) for ReduceOnPlateauParamScheduler

* [Docs] add docs for rule

* [Fix] fix pop from empty list bug in test

* [Fix] fix check param_schedulers is not built bug

* [Fix] fix step_args bug and without runner._train_loop bug

* [Fix] fix step_args bug and without runner._train_loop bug

* [Fix] fix scheduler type bug

* [Test] rename step_args to step_kwargs

* [Fix] remove redundancy check

* [Test] remove redundancy check

* Apply suggestions from code review

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* [Test] fix some defects

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>
Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* fix zero_optimizer error with param groups when pytorch < 1.12.0

* add docstring

* fix docstring

* add unittest

* change ut to use a valid paramwise_cfg

* modify ut

* fix as comments

* [Fix] Fix bias decay mult of depth-wise conv.

* support flatten param weight decay multiplier

* add unit tests

* REMOVE TODO

* update doc

* [Docs] Fix docstring format

* minor refine

* minor refine

* [Enhance]

Support step arugments and zero arguments with update_params

* Update mmengine/optim/optimizer/optimizer_wrapper.py

* Update mmengine/optim/optimizer/optimizer_wrapper.py

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>
Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* [Enhance] support eta_min_ratio in CosineAnnealingParamScheduler

* [doc] fix docstring

* [Enhance] add ut for eta_min_ratio

* [doc] update docstring

* avoid bc-breaking of eta_min

* [doc] add docstring in CosineAnnealingParamScheduler and CosineAnnealingMomentum

* Apply suggestions from code review

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* Replace print with print_log for those function called by runner

* minor refine

* Fix as comment

* [Fix] fix CosineRestart eta_min

* add ut case

* Enhance unit test

Enhance unit test

* remove unused code

Co-authored-by: HAOCHENYE <21724054@zju.edu.cn>

* [Feature] Support torch ZeRORedundancyOptimizer

Co-authored-by: Junhwa Song <ethan9867@gmail.com>
Signed-off-by: Junhwa Song <ethan9867@gmail.com>
Signed-off-by: Hakjin Lee <nijkah@gmail.com>

* lint

* Fix saving optimizer state_dict

* Fix handling import error

* Add test case

* fix UT

* Revert "fix UT"

This reverts commit dd64538960ff7440c6020f533d43945ffc23f2d2.

* fix handling import in UT

* Fix saving zero checkpoint and delete redundant master_only

* lint

* test unittest

* Fix handling impor error

* Fix UT condition

* Edit docstrings

* Fix typo

* Skip redundant procudure in checkpoint hook

* fix typo again

* Update mmengine/optim/optimizer/zero_optimizer.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* Add api info

* lint

* Fix lint

* Handling AmpOptimWrapper case

* handling overlap_with_ddp

* Fix error

Signed-off-by: Junhwa Song <ethan9867@gmail.com>
Signed-off-by: Hakjin Lee <nijkah@gmail.com>
Co-authored-by: Junhwa Song <ethan9867@gmail.com>
Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

Fix typo

* init npu

* Update mmengine/optim/optimizer/amp_optimizer_wrapper.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* Update mmengine/dist/dist.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* change to is_hccl_backend

* Update mmengine/optim/optimizer/amp_optimizer_wrapper.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* add comment with AmpOptimWrapper

* Update mmengine/runner/amp.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* Update mmengine/runner/amp.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* add npu fn in base_model

* Update mmengine/optim/optimizer/amp_optimizer_wrapper.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* clean lint

* Update mmengine/optim/optimizer/amp_optimizer_wrapper.py

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>

* Update mmengine/model/base_model/base_model.py

Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>

* add is_npu_available

* try to fix

* Add comments

* Refine grammar

Co-authored-by: Zaida Zhou <58739961+zhouzaida@users.noreply.github.com>
Co-authored-by: Mashiro <57566630+HAOCHENYE@users.noreply.github.com>
Co-authored-by: HAOCHENYE <21724054@zju.edu.cn>

* [Enhance] add documents for , and support clip grad by value

* refine docstring

* fix as comment

* Fix as comment

* minor refine

* minor refine

* remove error comment for clip grad

* refine docstring

* Upgrade the versions of pre-commit-hooks

* update zh-cn.yaml

* add args to OptimWrapper.step backward zero_grad

* minor refine

* minor refine

* Rename data to structure

* adjust the way to import module

* adjust the way to import module

* rename Structure to Data Structures in docs api

* rename structure to structures

* support using some modules of mmengine without torch

* fix circleci config

* fix circleci config

* fix registry ut

* minor fix

* move init method from model/utils to model/weight_init.py

* move init method from model/utils to model/weight_init.py

* move sync_bn to model

* move functions depending on torch to dl_utils

* format import

* fix logging ut

* add weight init in model/__init__.py

* move get_config and get_model to mmengine/hub

* move log_processor.py to mmengine/runner

* fix ut

* Add TimeCounter in dl_utils/__init__.py

* [Feature] Support CosineRestartParamScheduler.

* add ut and docstring

* add docstring

* fix save scheduler state dict with optim wrapper

* remove for loop and inherit TestParameterScheduler

* remove for loop and inherit TestParameterScheduler

* minor refine

* [Enhance] Auto set the `end` of param schedulers.

* Add log output and unit test

* Update docstring

* Update unit tests of `CosineAnnealingParamScheduler`.

* fix optimizer wrapper counts

* fix ut

* [Enhance] Support scheduling betas with MomentumScheduler.

* enhance ut

* test adam betas

* enhance ut

* enhance ut

* add autocast wrapper

* fix docstring

* fix docstring

* fix compare version

* fix unit test

* fix incompatible arguments

* fix as comment

* fix unit test

* rename auto_cast to autocast

* merge context

* update unit test

* add docstring

* fix bug in AmpOptimWrapper

* add docstring for backward

* add warning and docstring for accumuate gradient

* fix docstring

* fix docstring

* add params_group method

* fix as comment

* fix as comment

* make default_value of loss_scale to dynamic

* Fix docstring

* decouple should update and should no sync

* rename attribute in OptimWrapper

* fix docstring

* fix comment

* fix comment

* fix as comment

* fix as comment and add unit test

* [Features]Add OnecycleLR

* [Features]Add OnecycleLR

* yapf disable

* build_iter_from_epoch

* add epoch

* fix args

* fix according to comments;

* lr-param

* fix according to comments

* defaults -> default to

* remove epoch and steps per step

* variabel names

* fix build optimizer wrapper without type

* refine logic

* fix as comment

* fix optim_wrapper config error in docstring and unit test

* refine docstring of build_optim_wrapper

* Support multiple optimizers

* minor refinement

* improve unit tests

* minor fix

* Update unit tests for resuming or saving ckpt for multiple optimizers

* refine docstring

* refine docstring

* fix typo

* update docstring

* refactor the logic to build multiple optimizers

* resolve comments

* ParamSchedulers spports multiple optimizers

* add optimizer_wrapper

* fix comment and docstirng

* fix unit test

* add unit test

* refine docstring

* RuntimeInfoHook supports printing multi learning rates

* resolve comments

* add optimizer_wrapper

* fix mypy

* fix lint

* fix OptimizerWrapperDict docstring and add unit test

* rename OptimizerWrapper to OptimWrapper, OptimWrapperDict inherit OptimWrapper, and fix as comment

* Fix AmpOptimizerWrapper

* rename build_optmizer_wrapper to build_optim_wrapper

* refine optimizer wrapper

* fix AmpOptimWrapper.step, docstring

* resolve confict

* rename DefaultOptimConstructor

* fix as comment

* rename clig grad auguments

* refactor optim_wrapper config

* fix docstring of DefaultOptimWrapperConstructor

fix docstring of DefaultOptimWrapperConstructor

* add get_lr method to OptimWrapper and OptimWrapperDict

* skip some amp unit test

* fix unit test

* fix get_lr, get_momentum docstring

* refactor get_lr, get_momentum, fix as comment

* fix error message

Co-authored-by: zhouzaida <zhouzaida@163.com>

* [feat]:support display paramwise result in constructing optimizer

* [fix]:fix format issue

* delete unnecessary rank logic and fix format

* [Feature] Support convert epoch-based schedulers to iter-based.

* Support convert and refactor LR and Momentum to mixin.

* Add unit tests

* fix args and add runner ut

* resolve comments

* improve the coverage of docstring

* update