Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

release memory of task node on the fly #4735

Open
wants to merge 4,485 commits into
base: master
Choose a base branch
from
Open

release memory of task node on the fly #4735

wants to merge 4,485 commits into from

Conversation

levi1993
Copy link
Contributor

No description provided.

hjchen2 and others added 30 commits March 24, 2021 10:43
@hjchen2 @clackhan @oneflow-ci-bot
[one::OpBuilder] op interpreter core (#4407)
1e129b6 
* del object_storage.cpp

* use name GLOBAL_PARA_SYM2SHARED_OPKENEL_OBJ_MUTEX

* mig CheckRefInBlobObjectParallelDesc and  OperandBlobObjects rel api

* mig _StatelessCall

* mig _StatelessCall

* [one::OpBuilder] Refactor Operation.

* mig StatelessCall api

* mig StatefulCall

* mig callback api

* mig MakeLazyRefBlobObject

* refactor CudaHostPinBlob

* sort out InstructionsBuilder api

* [one::OpBuilder] Refine

* mig PhysicalRun and LogicalRun

* use oneflow_api.deprecated.LogicalRun & PhysicalRun

* delete vm_util.py

* change FindOrCreateDelegateBlobObject args

* add SetShuttingDown

* rm python_interpreter_util.py

* add blank line

* mig BlobCache

* use FindOrCreateDelegateBlobObject in c++

* refactor session_util.cpp

* use IsShutDown

* refactor BlobRegister

* [one::OpBuilder] Refactor OpExpr and OpExprInterpreter.

* [one::OpBuilder] Remove member function evaluate.

* [one::OpBuilder] Remove OpInterpreter to facilitate CR.

* [one::OpBuilder] Refine

* fix distribute test exit bug

* del comment

* [one::OpBuilder] Add more op exprs.

* mig id_util and scope_util

* use cfg_op_conf and Object*

* use Object*

* del _

* fix func name error

* [one::OpBuilder] Refine

* [one::OpBuilder] Modify op input names by InOutBnAccessor.

* [one::OpBuilder] Fix op expr python api.

* use MapAt and shared_ptr

* [one::OpBuilder] Using indexed ibns and obns instead of tensor names.

* [one::OpBuilder] Update

* [one::OpBuilder] Complete op interpreter in the main.

* use shared_ptr or const ref

* minor fix

* add todo

* [one::OpBuilder] Export and extend BoxingUtil in python.

* [one::OpBuilder] Support variable op interpretation.

* [one::OpBuilder] Refine

* [one::OpBuilder] Refine placeholder prefix.

* minor fix

* minor djustment

* minor fix

* [one::OpBuilder] Migrate snapshot manager.

* minor optimize

* minor fix

* minor optimize

* [one::OpBuilder] Migrate Session.

* [one::OpBuilder] Fix return type.

* [one::OpBuilder] Refine variable interpretation.

* minor optimize

* fix bug

* [one::OpBuilder] Call python FeePath.

* [one::OpBuilder] Fix typo.

* minor fix

* [one::OpBuilder] Fix merge bugs.

* [one::OpBuilder] Bugfix.

* [one::OpBuilder] Refine

* [one::OpBuilder] Bugfix

* [one::OpBuilder] Refine

* [one::OpBuilder] Fix typo.

* [one::OpBuilder] Fix placeholder prefix of op builer.

* [one::OpBuilder] Set output blob object after running the instruction.

* [one::OpBuilder] Remove TensorNameScope temporarily.

* [one::OpBuilder] Fix api.

* [one::OpBuilder] Add TensorNameScope.

* [one::OpBuilder] Modify the op builder apis to return Maybe type.

* [one::OpBuilder] Remove unused header.

* [one::OpBuilder] Fix

* [one::OpBuilder] Fix typo

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Create tensor from blob object.

* [one::OpBuilder] Return Maybe in op interpreter.

* [one::OpBuilder] Fix merge conflicts and reformat.

* [one::OpBuilder] Remove unused code.

* [one::OpBuilder] Go through eager mode.

* [one::OpBuilder] Fix typo

* [one::OpBuilder] Refine

* [one::OpBuilder] Remove redundant file system and refine code style.

* [one::OpBuilder] Use TensorTuple instead of TensorList and provide method to access the valide interpreter.

* [one::OpBuilder] Fix bug

* [one::OpBuilder] Remove state input.

* [one::OpBuilder] Create output tensors for lazy mode.

* [one::OpBuilder] Change pybind hold type to shared_ptr for Callback, Watch and BoxingUtil.

* [one::OpBuilder] Move to deprecated.

* [one::OpBuilder] Remove unused code.

* [one::OpBuilder] Fix typo.

* [one::OpBuilder] Support TensorTuple input and fix typo.

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Change pybind hold type to shared_ptr for Callback, Watch and BoxingUtil.

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Migrate snapshot manager.

* [one::OpBuilder] Remove redundant file system and refine code style.

* [one::OpBuilder] Migrate Session.

* [one::OpBuilder] Refine

* [one::OpBuilder] Refine

* [one::OpBuilder] Call python FeePath.

* [one::OpBuilder] Export and extend BoxingUtil in python.

* [one::OpBuilder] Deprecate boxing util python api.

* [one::OpBuilder] Reformat

* Merge

* Refine

* Use reference op expr.

* Fix default value

* Fix LocalFS use bug

* Fix and refine

* Return Maybe

* Reformat

* Refine

* Remove initialization to simplify variable op interpretation.

* Revert SnapshotManager and Session

* Fix register BoxingUtil

* Remove unused header

* Revert foreign callback

* Refine code style

* Fix register BoxingUtil

* Get device from parallel desc to make a new tensor.

* Boxing call instructions and refine.

* Fix ci error

* Swap the op conf since the move constructor was not provided by protobuf messages.

* Refine

* Refine code style

* Reformat

* Fix error merge

* Make interpreter stateless and remove OpExprInterpContext.

* Refine

* reformat

* Fix typo

* Revert interpreter interface

* Use move assignment since protobuf has been upgrade.

* Refine

* Refine

Co-authored-by: clackhan <han_binbin@163.com>
Co-authored-by: hjchen2 <hjchen2>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 8c7476c
@jackalcooper
Refine label check of PR (#4486)
da24db6 
* refine

* fix yml syntax

* use pr.*

* refine

Former-commit-id: ad4dbf3
@jackalcooper @oneflow-ci-bot
Add XLA CI back (#4485)
8fcb985 
* add xla back

* allow fail

* refine

* refine

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: b2c5212
@guo-ran @oneflow-ci-bot
interface_op support parallel_distribution (#4479)
76ecef1 
* interface_op support parallel_distribution

* add JUST

* fix

* fix

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 9d6ab3f
@jackalcooper @oneflow-ci-bot
Check xla is enable in script (#4490)
de2d579 
* refine

* add check

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 031dd54
@poohRui @lixinqi @oneflow-ci-bot
Refactor builder instr pb list to instr msg list (#4487)
2b26de2 
* refactor_builder_instr_pb_list_to_instr_msg_list todo list

* modified intruction proto list to msg list

* remove header to cpp, add cfg constructor

* modified variable life-time, using template for instruction msg and operand

* fix bug in instruction builder construction

* add cfg and template init for operand

* fix id to symbol

* fix function parameter name

* fix template link error

* format

Co-authored-by: lixinqi <lixinqi0703106@163.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: eab2f0d
@chengtbf @oneflow-ci-bot
Remove PackedBlob and packed_id (#4492)
7f3a525 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 14d8c25
@wyg1997 @oneflow-ci-bot
Feat autograd mode (#4496)
3d4b3d8 
* feat(AutoGradMode): add AutoGradMode

* style(*): fix typo

* feat(AutoGradMode): refine codes

* style(*): refine codes

* style(*): refine codes

* style(*): use namespace instead of static

* style(AutoGradMode): use class instead of struct

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 9a11c3f
@jackalcooper @clackhan @oneflow-ci-bot
Use local rpc backend by default if single process (#4493)
0ae13ad 
* always build local rpc backend

* only use grpc if it is multi process or multi machine

* rm py35

* add flags to enable

Co-authored-by: binbinHan <han_binbin@163.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: dbcc369
@chengtbf @oneflow-ci-bot
fixbug: mem reuse ctrl safe guard when block has multi-sink (#4497)
acd52f3 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 306e0e9
@mosout @oneflow-ci-bot
add some quantization ops conversion in onnx (#4465)
2a65ed1 
* add min_max_observer and moving_average_min_max_observer conversion in onnx

* fix op's name

* fix moving_average_min_max_observer

* update quantization ops conversion

* update quantization ops conversion and tests

* update ops version

* delete auto-imported package

* update min_max_observer op

* format

* fix test_quantization_aware_training

* update test_quantize_op

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: f5eb0fa
@guo-ran @oneflow-ci-bot
user_kernel support parallel_distribution (#4489)
308fa02 
* user_kernel support parallel_distribution

* refine

* fix

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 0f4870f
@poohRui @lixinqi @oneflow-ci-bot
construct pure intr msg list without pb (#4494)
6ae8b0e 
* refactor_builder_instr_pb_list_to_instr_msg_list todo list

* modified intruction proto list to msg list

* remove header to cpp, add cfg constructor

* modified variable life-time, using template for instruction msg and operand

* fix bug in instruction builder construction

* add cfg and template init for operand

* fix id to symbol

* fix function parameter name

* fix template link error

* format

* replace mirror

* broadcast object

* build send instruction

* build recv instruction

* cuda host

* lazy reference

* remain instruction

* minor fix

* minor fix

* minor fix

* minor fix

* add_del_object_operand

* fix typo

* add delete dependent instructions

* fix mutable operand error

* remove useless functions

* add ignore instruction function

* fix wrong use for intr_msg_list constructor

Co-authored-by: lixinqi <lixinqi0703106@163.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 4c26152
@chengtbf
Remove RegstDesc::Lock() and TaskNode::LockRegst() (#4506)
d202344 
Former-commit-id: a7b53ed
@jackalcooper @oneflow-ci-bot
Always enable local rpc backend (#4505)
ce985ea 
* always enable local

* add static_assert

* add log info

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 54d5520
@chengtbf @oneflow-ci-bot
Remove CtrlRegstDesc::returned_regst_num and reliant_regst_desc_id (#…
57ec3bd 
…4507)

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: b564020
@guo-ran @liujuncheng
CreateOpKernelState return nullptr when parallel_num = 1 (#4511)
7d44e93 
Co-authored-by: Juncheng <liujuncheng1022@gmail.com>
Former-commit-id: 640084e
@lixinqi
add sequantial callback instruction (#4503)
e7739d3 
* add sequantial callback instruction

* add a test_case for sequential instruction type

* refactor RunLogicalInstruction/RunPhysicalInstruction

* refactor RunLogicalInstruction/RunPhysicalInstruction

Former-commit-id: a6d0307
@guo-ran @oneflow-ci-bot
Hierarchical boxing sub graph (#4500)
3c08cb6 
* hierarchical boxing sub_graph

* refine

* fix

* refine

* refine

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 3d84630
@clackhan @oneflow-ci-bot
handle ctrl msg from other rank (#4491)
2a404cd 
* handle ctrl msg from other rank

* static way

* use check

* add todo

* add CHECK

* add DumpToConsumedRegstDescId2Addr function

* fix comment

* handle returned_regst_num

* optimize code

* rename regst_desc_id2regst_desc_addr_

* fix Segfault fault

* remove returned_regst_num

* rename arg and function

* add info im plan

* use name producer_task_id

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 4380494
@hjchen2 @clackhan @oneflow-ci-bot
[one::OpBuilder] Contruct variable and user op from Python (#4458)
60c55b4 
* use symbol::Storage<OperatorConfSymbol>

* _NewOpKernelObject

* mig OpKernelObject

* mig object_storage

* make of_format

* del comment

* std::function<void(Object*)

* mig NewOpKernelObject and _StatefulCallOpKernel

* mig _StatefulCallOpKernel and GetSharedOpKernelObject4ParallelConfSymbol

* del object_storage.cpp

* use name GLOBAL_PARA_SYM2SHARED_OPKENEL_OBJ_MUTEX

* mig CheckRefInBlobObjectParallelDesc and  OperandBlobObjects rel api

* mig _StatelessCall

* mig _StatelessCall

* [one::OpBuilder] Refactor Operation.

* mig StatelessCall api

* mig StatefulCall

* mig callback api

* mig MakeLazyRefBlobObject

* refactor CudaHostPinBlob

* sort out InstructionsBuilder api

* [one::OpBuilder] Refine

* mig PhysicalRun and LogicalRun

* use oneflow_api.deprecated.LogicalRun & PhysicalRun

* delete vm_util.py

* change FindOrCreateDelegateBlobObject args

* add SetShuttingDown

* rm python_interpreter_util.py

* add blank line

* mig BlobCache

* use FindOrCreateDelegateBlobObject in c++

* refactor session_util.cpp

* use IsShutDown

* refactor BlobRegister

* [one::OpBuilder] Refactor OpExpr and OpExprInterpreter.

* [one::OpBuilder] Remove member function evaluate.

* [one::OpBuilder] Remove OpInterpreter to facilitate CR.

* [one::OpBuilder] Refine

* fix distribute test exit bug

* del comment

* [one::OpBuilder] Add more op exprs.

* mig id_util and scope_util

* use cfg_op_conf and Object*

* use Object*

* del _

* fix func name error

* [one::OpBuilder] Refine

* [one::OpBuilder] Modify op input names by InOutBnAccessor.

* [one::OpBuilder] Fix op expr python api.

* use MapAt and shared_ptr

* [one::OpBuilder] Using indexed ibns and obns instead of tensor names.

* [one::OpBuilder] Update

* [one::OpBuilder] Complete op interpreter in the main.

* use shared_ptr or const ref

* minor fix

* add todo

* [one::OpBuilder] Export and extend BoxingUtil in python.

* [one::OpBuilder] Support variable op interpretation.

* [one::OpBuilder] Refine

* [one::OpBuilder] Refine placeholder prefix.

* minor fix

* minor djustment

* minor fix

* [one::OpBuilder] Migrate snapshot manager.

* minor optimize

* minor fix

* minor optimize

* [one::OpBuilder] Migrate Session.

* [one::OpBuilder] Fix return type.

* [one::OpBuilder] Refine variable interpretation.

* minor optimize

* fix bug

* [one::OpBuilder] Call python FeePath.

* [one::OpBuilder] Fix typo.

* minor fix

* [one::OpBuilder] Fix merge bugs.

* [one::OpBuilder] Bugfix.

* [one::OpBuilder] Refine

* [one::OpBuilder] Bugfix

* [one::OpBuilder] Refine

* [one::OpBuilder] Fix typo.

* [one::OpBuilder] Fix placeholder prefix of op builer.

* [one::OpBuilder] Set output blob object after running the instruction.

* [one::OpBuilder] Remove TensorNameScope temporarily.

* [one::OpBuilder] Fix api.

* [one::OpBuilder] Add TensorNameScope.

* [one::OpBuilder] Modify the op builder apis to return Maybe type.

* [one::OpBuilder] Remove unused header.

* [one::OpBuilder] Fix

* [one::OpBuilder] Fix typo

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Create tensor from blob object.

* [one::OpBuilder] Return Maybe in op interpreter.

* [one::OpBuilder] Fix merge conflicts and reformat.

* [one::OpBuilder] Remove unused code.

* [one::OpBuilder] Go through eager mode.

* [one::OpBuilder] Fix typo

* [one::OpBuilder] Refine

* [one::OpBuilder] Remove redundant file system and refine code style.

* [one::OpBuilder] Use TensorTuple instead of TensorList and provide method to access the valide interpreter.

* [one::OpBuilder] Fix bug

* [one::OpBuilder] Remove state input.

* [one::OpBuilder] Create output tensors for lazy mode.

* [one::OpBuilder] Change pybind hold type to shared_ptr for Callback, Watch and BoxingUtil.

* [one::OpBuilder] Move to deprecated.

* [one::OpBuilder] Remove unused code.

* [one::OpBuilder] Fix typo.

* [one::OpBuilder] Support TensorTuple input and fix typo.

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Change pybind hold type to shared_ptr for Callback, Watch and BoxingUtil.

* [one::OpBuilder] Fix bugs.

* [one::OpBuilder] Migrate snapshot manager.

* [one::OpBuilder] Remove redundant file system and refine code style.

* [one::OpBuilder] Migrate Session.

* [one::OpBuilder] Refine

* [one::OpBuilder] Refine

* [one::OpBuilder] Call python FeePath.

* [one::OpBuilder] Export and extend BoxingUtil in python.

* [one::OpBuilder] Deprecate boxing util python api.

* [one::OpBuilder] Reformat

* Merge

* Refine

* Use reference op expr.

* Fix default value

* Fix LocalFS use bug

* Fix and refine

* Return Maybe

* Reformat

* Refine

* Remove initialization to simplify variable op interpretation.

* Revert SnapshotManager and Session

* Fix register BoxingUtil

* Remove unused header

* Revert foreign callback

* Refine code style

* Fix register BoxingUtil

* Get device from parallel desc to make a new tensor.

* Boxing call instructions and refine.

* Boxing call instructions and refine.

* Construct variable and user op from Python.

* Construct variable and user op from Python.

* Fix ci error

* Swap the op conf since the move constructor was not provided by protobuf messages.

* Refine

* Refine code style

* Reformat

* Fix error merge

* Make interpreter stateless and remove OpExprInterpContext.

* Refine

* reformat

* Refine

Co-authored-by: clackhan <han_binbin@163.com>
Co-authored-by: hjchen2 <hjchen2>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 1117632
@guo-ran
operator infer parallel_distribution (#4519)
3205750 
* operator infer parallel_distribution

Former-commit-id: 82b6846
@lixinqi @oneflow-ci-bot
Sequantial instruction (#4521)
0b75935 
* add sequantial callback instruction

* add a test_case for sequential instruction type

* refactor RunLogicalInstruction/RunPhysicalInstruction

* refactor RunLogicalInstruction/RunPhysicalInstruction

* refactor front sequential instruction

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: bb4a310
@poohRui @oneflow-ci-bot
fix tensor construtor (#4522)
8e8d2ad 
* fix construtor

* fix construtor

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 54d069f
@guo-ran
indexed_slices_optimizer_rewrite_pass parallel_cast to hierarchical_p…
71c0337 
…arallel_cast (#4523)

Former-commit-id: 3b5cef7
@mosout @oneflow-ci-bot
add fake_quantization op conversion in onnx (#4512)
aa67c5f 
* add min_max_observer and moving_average_min_max_observer conversion in onnx

* fix op's name

* fix moving_average_min_max_observer

* update quantization ops conversion

* update quantization ops conversion and tests

* update ops version

* delete auto-imported package

* update min_max_observer op

* format

* fix test_quantization_aware_training

* update test_quantize_op

* add fake_quantization conversion in onnx

* format quantize.py

* fix fake_quantization

* update fake_quantization conversion

* update fake_quantization conversion and its test

* update quantization_aware_training

* update quantization_aware_training

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 1fef03a
@clackhan
Fix compiler bug of multi node with rank info bootstrap (#4524)
9eadf06 
* handle ctrl msg from other rank

* static way

* use check

* add todo

* add CHECK

* add DumpToConsumedRegstDescId2Addr function

* fix comment

* handle returned_regst_num

* optimize code

* rename regst_desc_id2regst_desc_addr_

* fix Segfault fault

* remove returned_regst_num

* rename arg and function

* add info im plan

* use name producer_task_id

* fix_bug_of_multi_node_with_rank_info_bootstrap

* fix bug

Former-commit-id: 2fe1b28
@liujuncheng @oneflow-ci-bot
ONEFLOW_PROFILER_KERNEL_PROFILE_KERNEL_FORWARD_RANGE (#4525)
4516679 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 30f3439
@wyg1997
Feat autograd engine patch (#4529)
b65f9e4 
* feat(*): implement autograd engine

* feat(OpExprHelper): add op_expr_helper codes

* style(*): remove outdated TODO

* style(*): refine check message

Former-commit-id: 74d7a71
@strint
oneflow.Model for function style api (#4246)
ee2f97d 
* trainer structure

* add test

* add nnmodel api

* nn Model draft

* try run global_func in Model

* fit to be refined

* model run global_func train & eval

* nn Model for function style execution draf test pass

* refactor nn model

* nn model with nessary component

* format

* rm nn prefix of Model

* flow.Model multi-task numpy-input

* (flow.Model)op_dataload support multi job

* (flow.Model) auto job_func signature for numpy input

* (flow.Model)support auto numpy input job

* (flow.Model) nump input multi job train test pass

* (flow.Model)fix classmethod

* fix test

* (oneflow.Model)training_step multi output, refine according to pep8

* (oneflow.Model)pep8 check pass by flake8

* Model refine

* Model fix typo

* oneflow.Model optimizer variable lazy get, numpy job signature to DataModule

* oneflow.Model merge and format

* oneflow.Model: comment empty func to be overried

* Optimizer: lazy get var add check and tips

* oneflow.Model: refactor

* oneflow.Model: refactor 2

* oneflow.Model: ModelStage -> SubStep, TrainStage -> TrainStep

* fix format

* oneflow.Model: SubStep to SubModel

* oneflow.Model: infer_oneflow_data_placeholder and _infer_job_signature

* add todo for GetCurrentJobName()

* fix typo

* oneflow.Model: refine error message

* fix format

* oneflow.Model: rm FunctionConfig in Model

* oneflow.Model config_exe to config_execution

* oneflow.Model: merge module

* Optimizer: user mode to confirm that Optimizer.Variable() is called inside a job

* simplify test

* oneflow.Model fix according to review

* rm useless code

Former-commit-id: 4de3595
@levi1993 levi1993 changed the base branch from master to lml/mem_optimize April 28, 2021 08:32
doombeaker and others added 3 commits April 28, 2021 16:57
@doombeaker @daquexian @oneflow-ci-bot
Add nn.AvgPool2d Module (#4623)
0beac81 
* need to be reformat

* reformat

* add docstring and refine test case

* add test case

* refine according to comments of wyg

* refine

* add TODO for asymmetric padding

Signed-off-by: daquexian <daquexian566@gmail.com>

Co-authored-by: daquexian <daquexian566@gmail.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: dfcd1d7
@BBuf @oneflow-ci-bot
add exp_tanh_gelu module (#4751)
6c942cc 
* add exp_tanh_gelu module

* fix comment

* fix comment

* fix comment

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 4b876e4
@levi1993 @jackalcooper @oneflow-ci-bot
Lml/mem optimize (#4725)
6f3fc5a 
* add memory detect info

* small fix in opattrref optimize

* use bitset

* refactor using vector

* refine

* refine

* rename

* refine

* address review

* address review

* refine

* refine

* address review

* smaller BITSET_SIZE

* refine

* refine

* refine

* refine nameing

* refine

* refine

* refine

* update

* delete swp file

* small update

* format fix

* format modify

* format modify

* Update compiler.cpp

fix for comment

* Update reshape_user_op_util.cpp

bug about reshape is fixed

Co-authored-by: jackalcooper <jackalcooper@gmail.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 3f728c0
Base automatically changed from lml/mem_optimize to master April 28, 2021 11:27
levi1993 and others added 9 commits April 28, 2021 19:38
@levi1993
Merge branch 'master' into lml/release_memory_of_task_node_on_the_fly
b1907da 
Former-commit-id: 1832d71
@MARD1NO @oneflow-ci-bot
add default value for oneflow range (#4771)
346c697 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 25f9f17
@guo-ran @oneflow-ci-bot @chengtbf
b21 boxing add ctrl_edge (#4770)
a63b53e 
* b21 boxing add ctrl_edge

* refine

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Co-authored-by: cheng cheng <472491134@qq.com>
Former-commit-id: a9f70c7
@guo-ran @oneflow-ci-bot
model_io_v2 process multi variable (#4762)
efa8dc7 
* model_io_v2 process multi variable

* refine

* reserve

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: f7b5bb0
@BBuf @daquexian @oneflow-ci-bot
add greater_less_argmax module (#4756)
c936361 
* add greater_less_argmax module

* fix comment

* add test_case

* fix conflict

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* fix comment

* format file

Co-authored-by: daquexian <daquexian566@gmail.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 4789392
@chengtbf @oneflow-ci-bot
Fix NcclLogicalS2S kernel comm BUG. (#4774)
f2b28e3 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 32a240c
@Ldpe2G @oneflow-ci-bot
Add flatten module and unit test (#4759)
87e83a6 
* add flatten module and unit test

* add changes according to review

* fix docs

* use nn.init

* add modification according to review

* remove useless code

* add more tests & fix doc

* fix default parameters for module and function

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 7940841
@chengtbf @guo-ran
@oneflow-ci-bot @liujuncheng
NCCL logical op support S1 to B (#4772)
ce04eff 
* NcclLogialOpAllGatherNoncontinuous S1 to B

* delete useless file

* fix user op attr err

* fix tmp_buffer

* Update oneflow/user/kernels/nccl_logical_kernels.cpp

* fix data size check in 2d s1-b

Co-authored-by: guo-ran <360112263@qq.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Co-authored-by: Juncheng <liujuncheng1022@gmail.com>
Former-commit-id: 92337ef
@levi1993
make ClearNodes and ClearEdges protected
d24969f 
Former-commit-id: 8d02926
levi1993
levi1993 commented Apr 29, 2021
View reviewed changes
oneflow/core/graph/task_graph.h Outdated Show resolved Hide resolved
oneflow/core/graph/graph.h Outdated Show resolved Hide resolved
MARD1NO and others added 4 commits April 29, 2021 11:44
@MARD1NO @oneflow-ci-bot
fix bug about pos_weight (#4768)
1266c94 
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 5af7d1b
@levi1993
rm prefix Protected
9361227 
Former-commit-id: 12f23d1
@levi1993
Merge branch 'lml/release_memory_of_task_node_on_the_fly' of github.c…
022a2eb 
…om:Oneflow-Inc/oneflow into lml/release_memory_of_task_node_on_the_fly

Former-commit-id: c42f43a
@levi1993
small fix
c0bcdc7 
Former-commit-id: 72c8e75
poohRui and others added 2 commits April 29, 2021 12:28
@poohRui @jackalcooper @oneflow-ci-bot
Create default log under log dir (#4766)
ba18214 
* move default log under log dir

* remove default env log from proto

* refine

Co-authored-by: Shenghang Tsai <jackalcooper@gmail.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: c402963
@hjchen2 @lixinqi @oneflow-ci-bot
Dev refactor attr value map (#4755)
4bc260a 
* Refactor

* Draft

* Refactor

* Add MutableCfgAttrValueMap

* implement AttrValueMap and ComposedAttrValueMap (#4767)

* Attr value util (#4773)

* implement AttrValueMap and ComposedAttrValueMap

* AttrValueUtil::ToProtoAttrValue

* Fix compile

* Fix compilation

* Rename AttrValueMap by AttrMap.

Co-authored-by: lixinqi <lixinqi0703106@163.com>
Co-authored-by: hjchen2 <hjchen2>
Co-authored-by: Li Xinqi <lixinqi2010@gmail.com>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
Former-commit-id: 354652e
@levi1993 levi1993 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot April 29, 2021 06:20
@CLAassistant
Copy link

CLAassistant commented Apr 29, 2021 •
edited
Loading

CLA assistant check
Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
9 out of 11 committers have signed the CLA.

✅ poohRui
✅ lixinqi
✅ leaves-zwx
✅ BBuf
✅ wanghongsheng01
✅ MARD1NO
✅ Ldpe2G
✅ hjchen2
✅ hengzi
❌ levi1993
❌ hsj0429
You have signed the CLA already but the status is still pending? Let us recheck it.

@jackalcooper jackalcooper force-pushed the lml/release_memory_of_task_node_on_the_fly branch from 04b328e to feb5d18 Compare June 2, 2021 07:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jackalcooper jackalcooper jackalcooper left review comments

@liujuncheng liujuncheng liujuncheng left review comments

@lixinqi lixinqi Awaiting requested review from lixinqi

@oneflow-ci-bot oneflow-ci-bot Awaiting requested review from oneflow-ci-bot

At least 1 approving review is required to merge this pull request.

Assignees
No one assigned
Labels
automerge enhancement system
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.