- 24 2月, 2021 7 次提交
-
-
由 Yurui Li 提交于
* initial tensor * combine blobdesc in tensor, initial MirrorTensor * fix typo * add consistent blob, tensorimpl and outer consistent tensor * add public method for tensor * remove unrelated file change * remove unrelated file change * fix: reviewer suggestions * fix: reviewer suggestions * Delete device.cpp * Add storage relative methods * Remove unchange file * Remove unchange file * add storage relative tensor interfaces * add some basic setter function * remove expose blob concept * remove useless header * remove useless header * add detail for new tensor implementation without blob desc * reoriganized * reoriganized * add get/set for blob object * refactor device remove is_lazy() * replace method in tensor * refactor device * refactor base tensor * remove header * refactor tensor impl * refactor tensor and device * refactor tensor impl * getters should return reference * add final * refactor tensor impl * add final for tensor impl * fix typo add #endif * unsolved comments * remove useless header * add protected method for tensor impl * compile and modified getters to return immutable parameters * code format * carefully deal with const property * remove constness of blob object, rename parallel_conf to parallel_desc Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* disable_group_boxing and change nccl logical order to dst * remove note * both support insert nccl logical ops as close as possible to Src/Dst node Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* Remove batch_axis * rm partial tick * temp fix bug of flow.math.add * fix bug of input_blob_def.split_axis init * [KEY] refine infer sbp order value consider logical shape enable split * ignore op get sbp sign ERROR. * filter and check valid sbp sign by logical shape; rm magic num * fix bug of return * merge rm sigmoid cross entropy op * rm sigmoid batch axis fn * more debug log for check valid in get sbp sign * rm useless check Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Shenghang Tsai 提交于
Co-authored-by: Ntsai <caishenghang@oneflow.org>
-
由 guo ran 提交于
Co-authored-by: Ncheng cheng <472491134@qq.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 qq_22305325 提交于
* add CtrlConf Proto * add HostListBootStrapClient * add HostListBootStrapServer * del OfOnceCall in host_list_boot_strap_client * add BootStrapServer/Client * Update control.proto del rank2ctrl_addr * add InitConfFromEnvDesc * add log * optimize code * add CHECK * InitCtrlConfFromEnvDesc * del useless args def * CtrlBootstrap * minor fix * refactor CtrlServer/CtrlClient with ProcessCtx * RankInfoBootstrap * fix bug and optimize * minor optimize * minor fix * use WorkerProcessInfo * minor optimize Co-authored-by: Nlixinqi <lixinqi0703106@163.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 daquexian 提交于
* quantize aware training support cambricon quantize formula * support custom scope and cambricon backend Signed-off-by: Ndaquexian <daquexian566@gmail.com> * align with latest fake quant ops Signed-off-by: Ndaquexian <daquexian566@gmail.com> * align with latest observer ops Signed-off-by: Ndaquexian <daquexian566@gmail.com> * fix wrong moving_min_max_stop_update_after_iters type Signed-off-by: Ndaquexian <daquexian566@gmail.com> * quantize aware training support cambricon quantize formula Co-authored-by: NLdpe2G <liangdepeng@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 23 2月, 2021 8 次提交
-
-
由 Juncheng 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 daquexian 提交于
Signed-off-by: Ndaquexian <daquexian566@gmail.com>
-
由 daquexian 提交于
* only write auto gen header if different Signed-off-by: Ndaquexian <daquexian566@gmail.com> * remove unused code Signed-off-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 qq_22305325 提交于
* add CtrlConf Proto * add HostListBootStrapClient * add HostListBootStrapServer * del OfOnceCall in host_list_boot_strap_client * add BootStrapServer/Client * Update control.proto del rank2ctrl_addr * add InitConfFromEnvDesc * add log * optimize code * add CHECK * InitCtrlConfFromEnvDesc * del useless args def * CtrlBootstrap * minor fix * refactor CtrlServer/CtrlClient with ProcessCtx Co-authored-by: Nlixinqi <lixinqi0703106@163.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Juncheng 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Juncheng 提交于
* Optimize RngTruncatedNormal * np.zeros -> np.empty, remove np.where in index Signed-off-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 OuYang Yu 提交于
* Migrate sigmoid_cross_entropy_with_logits to user op * add unittest * fix license * sigmoid cross entropy elemwise functor * add LabelType * remove KernelUtil::Mul * replace logf and expf with LogFunctor and ExpFunctor Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* remove deprecated checkpoint init code in ops unittest * retain deprecated api for session init in test_interface_op_read_and_write
-
- 22 2月, 2021 4 次提交
-
-
由 Shenghang Tsai 提交于
* higher TF_CPP_MIN_LOG_LEVEL * rm outdated tests * larger tol Co-authored-by: Ntsai <caishenghang@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 guo ran 提交于
* fix reduce_sum_like sbp * refine * rename Co-authored-by: Nguoran <guoran@oneflow.org>
-
由 qq_22305325 提交于
* add CtrlConf Proto * add HostListBootStrapClient * add HostListBootStrapServer * del OfOnceCall in host_list_boot_strap_client * add BootStrapServer/Client * Update control.proto del rank2ctrl_addr * add InitConfFromEnvDesc * add log * optimize code * add CHECK * InitCtrlConfFromEnvDesc * del useless args def * CtrlBootstrap * minor fix Co-authored-by: Nlixinqi <lixinqi0703106@163.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 doombeaker 提交于
* startup of device_registry, device_registry_mgr * refine(move GetDeviceString from devRegMgr to oneflow namespace) * Add some new Functions related to ToString. (#4185) * Add some new Functions related to ToString. * fix code according to review advices. * fp16 registration pseudocode (#4188) * pseudocode for demonstrating Signed-off-by: Ndaquexian <daquexian566@gmail.com> * GetZeroVal pseudocode Signed-off-by: Ndaquexian <daquexian566@gmail.com> * float16 resolved, GetZeroValue unresovled * refine GetZeroValue func * refactor float16 type uitls * remove useless core/framework/tensor.cpp Co-authored-by: Ndoombeaker <later@usopp.net> * move calling cudaRuntimeGetVersion to cpp * add device_register_gpu.cpp * refine device_register_gpu.cpp * try to fix cpu build error * fix cpu version bug * auto generated dev regs headers * remove exception in DeviceTag4Type/DeviceType4Tag temporarily * fix including bugs for python kernel test * refine according to review comments * reformat Co-authored-by: NWang Tuo <58287331+WangTuoxyty@users.noreply.github.com> Co-authored-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 21 2月, 2021 1 次提交
-
-
由 leaves-zwx 提交于
* test_style_transfer and refine inference session * fix bug * rm debug info * update old test * format * revert include header * refine ofrecord dataset * add FaceEmoreRecordDataset * test insightface save and load * format * change default model dir * rm useless code * add args for convenience of test * modify by review comment * support print stack for serving * new event loop when old one was closed * fix Co-authored-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: NTsai <caishenghang@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 20 2月, 2021 5 次提交
-
-
由 Yuchen Zhong 提交于
Signed-off-by: Nvycezhong <izhongyuchen@gmail.com> Co-authored-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 OuYang Yu 提交于
* remove PartialTick * remove op_conf.has_partial_tick_conf() Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* add NCCL_VERSION_CODE > 2700 macro in nccl logical kernels * refine macro position Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 leaves-zwx 提交于
* rm LocalWorkStreamId * rm AllocateLocalWorkStreamId in TaskNode * rm local work stream id in task node and commnet task node * rm local_work_stream_id param in NewTaskId * fix test Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 qq_22305325 提交于
* refactor CtrlClient * RpcServer * use Rpc * Update rpc_client.cpp add comment Co-authored-by: Nlixinqi <lixinqi0703106@163.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 19 2月, 2021 3 次提交
-
-
由 cheng cheng 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* Remove keep_header_only and BlobDesc::is_body_disabled * Remove InputBlobModifier::use_header_only and UserOps set_use_header_only
-
由 qq_22305325 提交于
* mig parallel_conf_util * mig BuildInitialScope BuildScopeWithNewParallelDesc BuildScopeWithNewParallelConf * add test of GetDeviceTagAndMachineDeviceIds * mig GetOpConfSymbol * fix BuildScopeWithNewParallelDesc input type error * use TRY * 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 * del comment * use cfg_op_conf and Object* * use Object* * del _ * fix func name error * Update object_storage.cpp rename mutex Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 18 2月, 2021 2 次提交
-
-
由 cheng cheng 提交于
* Enable insert nccl logical op pass * FindMaxConnectedSubgraphForGpuExecOrder~ * through order and interface * implement of insert nccl logical op in pass * add nccl logical op using UserOp Implement and EagerNcclCommMgr * add NCCL ReduceScatter op/kernel; refine pass impl of topo order * add NCCL logical op/kernel AllGather * fix bug of reduce scatter/ all gather infer shape * refine log and note * fix complier err build with CPU ONLY * support NCCL ALL2ALL and test pass of alexnet model parallel * rollback of diff in checkpointing_pass.cpp * rename to nccl_use_compute_stream; ResourceDesc::nccl_use_compute_stream; refine name for review; create nccl_comm_ in KernelCompute; * refine code for review * add unittest for nccl use compute stream * format test scripts * refine align
-
由 Juncheng 提交于
* Refactor InferBatchAxis * refine
-
- 17 2月, 2021 1 次提交
-
-
由 Li Xinqi 提交于
* source subset tick * remove useless header files * insert DstSubsetTickOp * remove incorrect CHECK * add tick op for each machine * TryBindBnWithOneofRegst * add sink tick op in main_job * refactor LinkMainJob * fix typo in task_graph * refactor AddGlobalCriticalSection * rename and refactor DstSubsetTick::InferBlobDescs and SrcSubsetTick::InferBlobDescs * add src_subset_tick for input-output critical section * refactor AutoSourceTick and AutoSinkTick * vectorizedly link main job * resize vectorh identity_tick_op_names then access elements * SrcSubsetTickCompTaskNode: bind bns and in_regst if bns is valid in current device * refactor optional input to repeated inputs for SrcSubsetTickOpConf * fix a bug in CaseCompTaskNode; fix a bug when create identity tick in main_job * 1) Insert tick between sourc tick and src_subset_tick; 2) Insert tick between dst_subset_tick and sink tick * stash code * refactor MakeMainJob by using Range::ForEachSubRange * refactor MakeMainJob by using Range::ForEachSubRange * rename ReentrantLockLinkPoint to ReentrantLockBackEdge * set piece id for regst sent by wait_and_send_ids actor * callback_notifier_sink_tick Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 15 2月, 2021 1 次提交
-
-
由 Li Xinqi 提交于
* source subset tick * remove useless header files * insert DstSubsetTickOp * remove incorrect CHECK * add tick op for each machine * TryBindBnWithOneofRegst * add sink tick op in main_job * refactor LinkMainJob * fix typo in task_graph * refactor AddGlobalCriticalSection * rename and refactor DstSubsetTick::InferBlobDescs and SrcSubsetTick::InferBlobDescs * add src_subset_tick for input-output critical section * refactor AutoSourceTick and AutoSinkTick * vectorizedly link main job * resize vectorh identity_tick_op_names then access elements * SrcSubsetTickCompTaskNode: bind bns and in_regst if bns is valid in current device * refactor optional input to repeated inputs for SrcSubsetTickOpConf * fix a bug in CaseCompTaskNode; fix a bug when create identity tick in main_job * 1) Insert tick between sourc tick and src_subset_tick; 2) Insert tick between dst_subset_tick and sink tick Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 14 2月, 2021 1 次提交
-
-
由 Li Xinqi 提交于
* source subset tick * remove useless header files * insert DstSubsetTickOp * remove incorrect CHECK * add tick op for each machine * TryBindBnWithOneofRegst * add sink tick op in main_job * refactor LinkMainJob * fix typo in task_graph * refactor AddGlobalCriticalSection * rename and refactor DstSubsetTick::InferBlobDescs and SrcSubsetTick::InferBlobDescs * add src_subset_tick for input-output critical section * refactor AutoSourceTick and AutoSinkTick * SrcSubsetTickCompTaskNode: bind bns and in_regst if bns is valid in current device * refactor optional input to repeated inputs for SrcSubsetTickOpConf Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 09 2月, 2021 5 次提交
-
-
由 Li Xinqi 提交于
* source subset tick * remove useless header files * insert DstSubsetTickOp * remove incorrect CHECK * add tick op for each machine * TryBindBnWithOneofRegst * fix typo in task_graph * rename and refactor DstSubsetTick::InferBlobDescs and SrcSubsetTick::InferBlobDescs * SrcSubsetTickCompTaskNode: bind bns and in_regst if bns is valid in current device * refactor optional input to repeated inputs for SrcSubsetTickOpConf Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 qq_22305325 提交于
* mig parallel_conf_util * mig BuildInitialScope BuildScopeWithNewParallelDesc BuildScopeWithNewParallelConf * add test of GetDeviceTagAndMachineDeviceIds * mig GetOpConfSymbol * fix BuildScopeWithNewParallelDesc input type error * use TRY * use symbol::Storage<OperatorConfSymbol> * _NewOpKernelObject * mig OpKernelObject * mig object_storage * make of_format * del comment * del comment * use cfg_op_conf and Object* * use Object*
-
由 OuYang Yu 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 qq_22305325 提交于
* mig parallel_conf_util * mig BuildInitialScope BuildScopeWithNewParallelDesc BuildScopeWithNewParallelConf * add test of GetDeviceTagAndMachineDeviceIds * mig GetOpConfSymbol * fix BuildScopeWithNewParallelDesc input type error * use TRY * use symbol::Storage<OperatorConfSymbol> * _NewOpKernelObject * del comment Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Zhenhua 提交于
* Add CTC Loss * Add backward kernel * Remove tf in test * Update api document * Add zero_infinity option * refine * Add 1n2d test case * Switch to consistent_view * Fix Eager mode * Remove duplicate license * Add grad check * Fix bw test * Fix bugs * Add op name * Refine * of_format * Expand annotation * Performance optimizing for cuda * Check input_length & target_lengths * Update __syncthreads Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 08 2月, 2021 2 次提交
-
-
由 ZZK 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Li Xinqi 提交于
* source subset tick * remove useless header files * insert DstSubsetTickOp * remove incorrect CHECK * TryBindBnWithOneofRegst * fix typo in task_graph * rename and refactor DstSubsetTick::InferBlobDescs and SrcSubsetTick::InferBlobDescs Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-