Co-authored-by: Ntsai <caishenghang@1f-dev.kbaeegfb1x0ubnoznzequyxzve.bx.internal.cloudapp.net>
Former-commit-id: 1e535df3

* add SetTrainConf api and some implementation of optimizer class

* set functionconfig type in constructor, refactor optimizer classes

* refactor to support lr_lbn, add Adam optimizer, update api_global_function

* make model_update_conf and primary_lr optional

* check model_update_conf and primary_lr

* add other lrschedulers and optimizers

* remove weight_decay, add AdamW, and other minor changes

* add ClipGradientConf

* refine

* refine

* PiecewiseScalingScheduler supports a single scale

* fix copy paste error

* rename val to clip_norm

* move grad_clipping to optimizer namespace, add WarmupConf

* check model_update_conf and primary_lr only when has_train_conf() is true

* support sequence loss, improve type check

* update optimizer in test_util.py

* replace str with Text in type annotations

* reformat

Former-commit-id: 7bc8cd72

* update readme

* python3 -m pip

* colon

* rm required

* To install dependencies, run:

Former-commit-id: e4e631e8

* fix script

* BUILD_RDMA default ON

* add rdma dependency in dockerfile

* fix rm tmp wheel

Former-commit-id: e33bd43d

* add dcgan test

* add 1n4c test for deconv

* multi-gpu result different

* fix merge

* add license

* format

* fix bn

* typo

* remove deconv 1n4c test
Co-authored-by: Njames plur <jamesplur7@gmail.com>
Former-commit-id: 128ccf70

* add dict return type of global function

* remove type judge of OriginFrom

* fix syntax err

* add test script

* fix code style

Former-commit-id: ef395840

1. replace default_distribute_strategy call by default_logical_view
2. fix WARNING info of deprecated default_distributr_strategy
Co-authored-by: NYao Chi <yaochi@oneflow.org>
Co-authored-by: NOuYang Yu <xuanjiuye@gmail.com>
Former-commit-id: c8363bcb

Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 5eaf879a

Co-authored-by: Ntsai <caishenghang@1f-dev.kbaeegfb1x0ubnoznzequyxzve.bx.internal.cloudapp.net>
Former-commit-id: 2b651408

Former-commit-id: 29e6c258

Former-commit-id: 12dfd815

Former-commit-id: ce98ed74

Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 80d8881e

Former-commit-id: 54f8ddda

* remove unused tmp_buffer_blob_object

* filter empty tmp_buffer blob

Former-commit-id: c5c3b783

* add docs to env_util.py function_util.py

* add docs to config_util.py

* add docs to session_util.py

* add docs to layer.py

* add docs to layers.upsample_2d

* add docs nn_ops

* format refine

Former-commit-id: ff780ff9

Former-commit-id: 560c50fc

Former-commit-id: 41047ac3

* add DeprecatedFixedTensorDef class

* add depreactedXXX function for MirroredTensorListDef and MirroredTensorDef

* remove docstring of deprecated functions
Co-authored-by: NYao Chi <yaochi@oneflow.org>
Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 0b66ea6e

Former-commit-id: 75807fc3

* oneflow.find_or_create_module

* fix eager coco_reader

* merge OpParallelAttribute and UpstreamSignature into OpNodeSignature

* OpNodeSignatureDesc

* OpNodeSignature

* remove duplicated api find_or_create_module

* fix OpNodeSignature gtest bugs

Former-commit-id: 959f18f4

* remove decode_ofrecord in test scripts; using new reader/decoder op

* fix for pass test

* benchmark runnable

* Add WARNING and assert not eager in decode ofrecord op

* fix transpose in cnn benchmark
Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 5da538ed

Co-authored-by: NYao Chi <yaochi@oneflow.org>
Former-commit-id: c7fd11cb

* remove_enable_eager_at_session

* fix code style

* fix unit test code

* Recover deleted test files

* conver EagerExecution to a normal class

Former-commit-id: 025a53d7

* oneflow.find_or_create_module

* fix eager coco_reader

Former-commit-id: 9fae2aab

Former-commit-id: f7df9642

* flow.scope.current_scope -> flow.current_scope

* reformat python

* add deprecated_cuurent_scope and WARNING info

* refomrat python
Co-authored-by: NYao Chi <yaochi@oneflow.org>
Former-commit-id: 58f788f2

Former-commit-id: 2bf6d602

* aadd manual trigger

* add back

* reorder

Former-commit-id: cd1be0fd

* fix push_util bugs

* fix coco op bugs

* UserOpModule

Former-commit-id: e7523a47

* refactor ParallelConf

* refactor parallel_conf

* fix parallel_conf init

* refactor opkernel_instruction_type init/compute

* fix test_cpu_only_user_op

* remove notes

* fux non_distributed_optimizer_pass.cpp bug

* fix code style

* fix a small bug
Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: dd0786dd

* create symbol link for deps

* naive script

* check in changes

* run script at root dir

* fix yum

* use python found by cmake to run black

* fix help

* fix requirements

* add ignore

* fix issue for python 3.5

* script working

* add comment

* quick fix

* update ci script

* quick fix

* quick fix

* refine script

* rm proxy before install mkl

* rm workaround

* rm dns

* fix docker it

* rm useless

* rm cd

* reorder

* fix exe test

* quick fix

* always clean files

* update setup py

* rm clean

* add house dir

* add trap

* dont retry when it

* add arg

* minor fix

* more reasonable default

* update install

* use proxy when run

* fmt

* fix condition

* reorder

* fix dir

* customized py dir

* use rel dir

* revert setup.py

* fix version

* fix ci

* fix mount

* fix case

* no proxy

* http proxy for https

* fix permission

* use env

* rm useless

* restore exe test

* fix format

* move onnx test since it is stable

* add clean sh

* configure GitHub Actions to kill the build when outdated

* test cancel

* rm if

* test cancel

* auto cancel doesn't work
Co-authored-by: Ntsai <caishenghang@oneflow.org>
Co-authored-by: Ntsai <caishenghang@1f-dev.kbaeegfb1x0ubnoznzequyxzve.bx.internal.cloudapp.net>
Former-commit-id: fb0affbe

* change default distribute_strategy to consistent view

* change Shape to shape

Former-commit-id: c70a9904

* port tf2onnx to oneflow and test on alexnet

* add flow_inputs to op decorator

* add unit tests and update many ops accordingly, fix broken optimizer, eliminate model_load_dir parameter by a trick

* fix problem that groups and dilation_rates are not used

* add conv tests for dilations and group

* disable the borken transpose optimizer

* support leaky_relu, clip_by_scalar op, mobilenetv2 model

* fix the broken transpose optimizer

* update weight initializers

* rename tf_op decorator to flow_op

* replace "tensorflow" with "oneflow", remove some tf-specific code, delete all rewriters since they are not needed for oneflow

* remove many ops not need to support now, add many tests

* remove duplicated file

* fix opset 11 pad op

* remove some unsupported ops, map add_n to sum op

* remove custom opset and controlflow ops

* add more ops and tests

* use PbMessage2TxtString

* remove unused code and polish

* rename utils.py to util.py

* remove unused parameters

* map distribute_* op to identity

* use ONEFLOW_DEBUG_MODE env var

* delete verbose_logging.py

* remove rewriter and rename some variables

* add onnx to install_requires

* update some comments

* add layernorm and format

* replace flow.function with flow.global_function

* refactor input_maps and output_maps, combine oneflow_to_onnx_naive and flowlist_to_onnx, remove layernorm

* reformat

* remove unused functions

* replace make_name and port_name with id_util.UniqueStr

* rename some function names and filenames

* rename flow_inputs->flow_ibns

* add onnx in dev-requirements.txt

* reformat

* remove 'target', refine comments

* in ci container, install python deps from dev-requirements.txt

* remove opset-1 converters if there is another converter for opset < 8

* rename some functions

* copy_dtype->CopyDtype

* check ENABLE_USER_OP != False instead of == True

* use absolute import

* support >2GB model

* Remove unused GraphUtil and 'PlaceHolderWithDefault'

* fix typo extrace->extract

* Add docs for flow.onnx.export

* Add typing hints

* wrap Export with session_ctx.try_init_default_session

* remove debug code in get_saved_tensor

* drop the support of gathernd opset < 11

* fix bug of equal/not_equal when opset < 11

* make default opset and tested opset both 10

* reformat

* optimize the import order

* update docs about default opset

* add assert to check the existence of model_save_dir

* remove oneflow.python dependency in tests, replace deprecated apis

* replace flow.deprecated.variable_scope to flow.scope.namespace

* run onnx export model test in ci

* add license

* fix typo

* run onnx test before op test

* fix onnx test script

Former-commit-id: 87740ae8

* fix test_categorical_ordinal_encoder

* reuse=False

* reuse=False in prelu

* update test

* update conv2d test and fix bug

* fix
Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 844fc266

Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com>
Former-commit-id: 0ba8fe20

Former-commit-id: b969303d

* fix push_util bugs

* fix coco op bugs

Former-commit-id: e01db73f

Former-commit-id: 7925cfcb

Former-commit-id: 2e333690