- 17 8月, 2021 11 次提交
-
-
由 Liang Depeng 提交于
* add clip_grad to optimizer * add clip grad unit test to sgd optimizer * add sgd graph clip grad test * refine * format * refine * fix * merge master * refine * refine Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Shijie 提交于
* where broadcast forward * delete py broadcast code * delete op_expr * add where backward broadcast gradient * scalar x op * add where scalar_x, scalar_y, scalar_xy ops & kernels * add where scalar_x, scalar_y, scalar_xy functors * fix bug * add where backward * add pybind api * add testcase * fix * fix 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
* Remove GlobalWorkStreamId/GlobalThrdId * refine Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Twice 提交于
Co-authored-by: NHoujiang Chen <chenhoujiangcug@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Kevin_Xiong 提交于
* add randperm with test and docs * format code * format * fix docs * format the code and add more tests * format code * Update test_randperm.py * Update randperm.py * add head * format codes * docs * Update test_randperm.py * Update test_randperm.py * add more tests * format * Update randperm.py * Update randperm_kernel.cu * Update randperm_kernel.cu * Update randperm_kernel.cpp * reconstruct the code * format the code * 2 * s * 1 * 1 * refine * add more test * refine code * fix according to comment * add more test Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Juncheng 提交于
-
由 Xiaoyu Xu 提交于
* lrs * prototype lrs * add cosine lr scheduler conf * add opt group * test opt group * add comment * add comment * add return for lr conf * check same optimizer in the opt group * check same optimizer in the opt group * check cosine lr compute * cosine lr_sch pass test * add_optimizer * rm outdated comment * make lr_sheduler optional and add test Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 ZZK 提交于
-
由 Xiaoyu Xu 提交于
* ref count del of session and env * auto format by CI * add test * pass graph in global test * auto format by CI * rm cuda * rm debug lof * rm debug info * move graph del sync from c to python * rm log * auto format by CI * refine test to pass ci * fix env sync call * auto format by CI * deal with graph destruction when graph not compiled * deal with not compiled graph destruction * print log * deal with single client sync * auto format by CI * rm py object ref * rm python obj ref count * add log of env * address comment * auto format by CI Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Yinggang Wang 提交于
* fix(*): fix inplace op circle reference bug * refine code * format Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 16 8月, 2021 11 次提交
-
-
由 ZZK 提交于
* process api dtype * process api tensor * small fix * fix to name func * still error * still error * revert master * fix symbol dtype part1 * import error * still error * still error * fix symbol * add Dtype4Datatype function * work in process * undefined symbol of tensorbuffer * macro bug * construction func return symbol dtype * process deprecated api * fix name * fix bug * change to use reference * fix format * fix name * small fix * fix symbol dtype * modify functor datatype to symbol dtype * fix to symbol dtype * add object as for symbol dtype * modify return symbol dtype * modify functor * modify autograd section * modify autograd * modify core * modify back for functional * modify functional impl * modify value types * modify value types * fix args * modify tensor * modify back * fix comment * small fix * auto format by CI * fix tensor buffer bug * modify new interface * modify new interface * remove annotation * auto format by CI * need refine * remove dtype4datatype * modify use Get * auto format by CI * small fix * auto format by CI * remove annotation * houjiang fix to symbol dtype pointer * fix dtype * fix format * remove annotation * small fix * add null ptr support * auto format by CI * use symbol dtype * fix * auto format by CI * small fix * fix in lazy op interpreter * auto format by CI * fix to use symbol dtype * small fix to use equal instead of is in single client * fix conflict * Format * Fix to use symbol dtype Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Ncheng cheng <472491134@qq.com>
-
由 Xiaoyu Zhang 提交于
* restruct sort op * add sort autotest * restruct argsort and add autotest * add tensor autotest * auto format by CI * add l1loss autotest * auto format by CI Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
* fix bug of SyncOnMasterFn atexit * move sync to func Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Houjiang Chen 提交于
* Sync access and assign indexing tensor. * Remove unused comments. Co-authored-by: binbinHan <han_binbin@163.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Xiaoyu Zhang 提交于
* try to fix prelu insightface bug * fix comment * auto format by CI Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Xiaoyu Zhang 提交于
* restruct logsoftmax and abs test * add hardtanh test * refine batchnorm autotest * add meshgrid autotest * add pow autotest * add stack autotest * delete prelu useless code * change Stack Module Test Api * fix comments * fix softmax bug * fix sign bug * fix sign * auto format by CI Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 l702572275 提交于
* add fmod * add grad * fotmat and delete comment * delete comment * delete comment * delete line * rename and format * merge master * format * fix error , add data type * add int8 ,delete f16 * modified mod example * auto format by CI Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org>
-
由 tingkuanpei 提交于
* Support combined_margin_loss op in flow.nn.modules * Follow review comment to modify * Follow review comment to modify * auto format by CI Co-authored-by: NYao Chi <later@usopp.net> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 GehangZhang 提交于
* fix conv1d conv2d conv3d useless args * fix showing bugs * add args for unsample * fix torch.float Co-authored-by: NYao Chi <later@usopp.net> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Zailiang 提交于
* functional.onehot added * functional api yaml updated * api yaml updated * add one_hot functional API * add Module * amend one_hot testcase * amend docstring * delete Module * amend review question * amend review question * amend flow.nn.functional.one_hot * amend functional init.py * delete one_hot import code * amend valueerror * auto format by CI * amend numclasses is -1 * add testcase * update test_one_hot * auto format by CI * amend one_hot param * amend docsting * amend onehot.py docstring * amend on_value and off_value * auto format by CI * amend docsting error * remove onehot * auto format by CI Co-authored-by: Ntangnana925 <85614052+tangnana925@users.noreply.github.com> Co-authored-by: Ntangnana <tnn_personal@163.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: MARD1NO <359521840@qq.com>
-
由 Houjiang Chen 提交于
* abstract InferSourceOpParallelDistribution * support specified device, add consistent arange. * format * Fix arange python api and unittest * Broadcast parallel default * Fix wrong merge * Fix error merge Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
- 15 8月, 2021 10 次提交
-
-
由 Juncheng 提交于
* Add cmake option USE_SYSTEM_NCCL nccl target * rm mark_as_advanced
-
由 Twice 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Shijie 提交于
* fix gather kernel 0 shape * recover module test Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Li Xinqi 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 ZZK 提交于
* add logical scalar kernel * add logical scalar op register * add functional api yaml * modify math functor * fix * reuse functor * fix * modify equal * modify greater * modify greater equal * modify less equal * modify less than * add not equal * modify not equal * fix format * remove partial sum * add newline * reuse base class * fix bin_op to binary_op * modify to Scalar * first restruct and anotate cuda * modify to no grad user op * restruct code and add dtype * export to pybind * remove redundant logic in python * bind python as false * remove annotation * fix dtype * support scalar in input or output * fix * Add magic method * add docs * auto format by CI * fix randn test * modify back * small fix * fix 0d tensor * auto format by CI * fix 0d test * auto format by CI * fix ddp bug Signed-off-by: Ndaquexian <daquexian566@gmail.com> * fix to use is * remove [0] in ddp.py Signed-off-by: Ndaquexian <daquexian566@gmail.com> * fix unittest * fix format * fix wrong unittest * skip free eager test * fix to use is not none Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Ndaquexian <daquexian566@gmail.com> Co-authored-by: Ncheng cheng <472491134@qq.com>
-
由 cheng cheng 提交于
* Fix BUG of LazyInterpret FreeEagerTensor memory shared with regst * remove note * remove debug
-
由 liufengwei0103 提交于
Co-authored-by: NZZK <42901638+MARD1NO@users.noreply.github.com>
-
由 Tianyu Zhao 提交于
* Rename `ParallelDistribution` to `NdSbp` * Rename `ParallelDistribution` to `NdSbp` * Rename `ParallelDistribution` to `NdSbp` * auto format by CI * Rename `ParallelDistribution` to `NdSbp` Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> 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 Shenghang Tsai 提交于
-
- 14 8月, 2021 8 次提交
-
-
由 Bowen Chen 提交于
* add flow.rand * update docstr * update docstr * add consistent_rand, add more tests * update random op * refine * refine, add range and int type to uniform_kernel * refine * refine * update doc * update doc * Refactor UniformDistribution * fix Co-authored-by: Nhjchen2 <chenhoujiangcug@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Li Xinqi 提交于
* SyncAccessBlobByCallback * refactor capture-by-reference to capture-by-value * refactor InstructionsBuilder::SyncAccessBlobByCallback Co-authored-by: NHoujiang Chen <chenhoujiangcug@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 cheng cheng 提交于
Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Shenghang Tsai 提交于
* cmake first class cuda support * refine * refien * refine * refein * refein * refeine * refine * refein * refine * refien * refgine * refien * refein * refein * rm useless * refien * refein * also link cuda libs if build static * refein * refien * add * Revert "add" This reverts commit d9e67ad1. * fix * refeine * retine Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-
由 Yinggang Wang 提交于
* feat(Tensor): support Tensor.__bool__() * test(Tensor): add tensor to bool test * docs(Tensor): refine is_nonzero document * format * fix(Tensor): fix Tensor.__bool___ bug * auto format by CI * fix(instancenorm): fix merge bug * fix(*): fix merge bugs Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Ncheng cheng <472491134@qq.com>
-
由 liufengwei0103 提交于
* refine code * refine code * optimize code * refine code * refine * back up * add tensor.to func * make of_format * remove to in pyTensor * sync gpu data * refine * refine * refine * refine * refine * refine * refine * refine * refine * backup * refine * rebase * check in gen py * merge master and fix bugs * address pr comments * eager boxing * address pr comments * fix b2p error * auto format by CI * remove boxing * export sbp * add tensor to_consistent * /minor fix * minor fix * refine * remove useless head file * Fix optional * remove to in tensor.cpp * update * Support symbol placement type in functional. * add sbp and sbp list arg * refine * use functional * refactor CastConsistentOpExpr * to_consistent(flow.B) backward * Cache op expr * add EagerNcclOpKernelState * refine * refine * refine * refine * refine * refine * minor fix * capture OpInterpContext * unimplemented apply * add GetNdSbp * add mutex * refine * merge EagerConsistentTensorImpl::NewWithPhyTensor and EagerConsistentTensorImpl::NewWithoutPhyTensor into EagerConsistentTensorImpl::New * rename functiona SyncData to SyncMetaAndData * fix function yml * refine * refine * refine collective boxing * make of_format * of_format * add to_local to pybind * refactor EagerBoxingInterpreter * minor fix * optimize CastParallelDistribution * add placement_sbp_util * minor fix * eager boxing backward * minor fix * sync shape and data when tensor_to_local * fix rpc_token bugs * fix p2s backward bug * refactor AsyncRpcCtx * set logical_shape correctly * simplify implementation of consistent_tensor.to_local * refine * initialize rpc_token with zero * refactor grad functions of to_consistent/to_local * refine * reformat and address pr comment * reformat * add check_meta_consistency in consistent2sonsistent * refactor eager_nccl_reduce lernel * refine * refine to_consistent api * ban_non_pod_data_in_eager_boxing * refine * refine * refine * backup code * THREAD_LOCAL_CACHED * Delete thread_local_cache.h * bugfix: DeviceId4ParallelId -> MachineId4ParallelId * optimize * support tensor str * Init code and can print consistent * refine format * remove useless to_consistent and format * refine code and print according data * attempt to support multi rank when fetch data * Revert "attempt to support multi rank when fetch data" This reverts commit ae56afad. * skip if tensor is consistent * delete useless * add comment * delete useless * traversal data to determine if int_mode * if consistent, return [...] * refine * add test and fix bug * add more assertTrue and delete useless * getitem using integer return scalar when tensor shape is [1] * add test cast * refine * fix spelling mistake * add op test and enhance in parse device * fix bug * fix docstr test bug and support to print meta * refine * auto format by CI * fix docstr in clip_grad.py * fix docstr * fix docstr and bug * the input shape parameter of reshape changed * add with flow.no_grad when operate tensor * fix docstr Co-authored-by: clackhan <han_binbin@163.com> Co-authored-by: Ntsai <jackalcooper@gmail.com> Co-authored-by: NXinqi Li <lixinqi0703106@163.com> Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Nhjchen2 <chenhoujiangcug@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: Nwyg1997 <wyg19970408@gmail.com> Co-authored-by: Ncheng cheng <472491134@qq.com>
-
由 leaves-zwx 提交于
* refine code * optimize code * refine code * refine * back up * add tensor.to func * make of_format * remove to in pyTensor * sync gpu data * refine * refine * refine * refine * refine * refine * refine * refine * refine * backup * refine * rebase * check in gen py * merge master and fix bugs * address pr comments * eager boxing * address pr comments * fix b2p error * auto format by CI * remove boxing * export sbp * add tensor to_consistent * /minor fix * minor fix * refine * remove useless head file * Fix optional * remove to in tensor.cpp * update * Support symbol placement type in functional. * add sbp and sbp list arg * refine * use functional * refactor CastConsistentOpExpr * to_consistent(flow.B) backward * Cache op expr * add EagerNcclOpKernelState * refine * refine * refine * refine * refine * refine * minor fix * capture OpInterpContext * unimplemented apply * add GetNdSbp * add mutex * refine * merge EagerConsistentTensorImpl::NewWithPhyTensor and EagerConsistentTensorImpl::NewWithoutPhyTensor into EagerConsistentTensorImpl::New * rename functiona SyncData to SyncMetaAndData * fix function yml * refine * refine * refine collective boxing * make of_format * of_format * add to_local to pybind * refactor EagerBoxingInterpreter * minor fix * optimize CastParallelDistribution * add placement_sbp_util * minor fix * eager boxing backward * minor fix * sync shape and data when tensor_to_local * fix rpc_token bugs * fix p2s backward bug * refactor AsyncRpcCtx * set logical_shape correctly * simplify implementation of consistent_tensor.to_local * refine * initialize rpc_token with zero * refactor grad functions of to_consistent/to_local * refine * reformat and address pr comment * reformat * add check_meta_consistency in consistent2sonsistent * refactor eager_nccl_reduce lernel * refine * refine to_consistent api * ban_non_pod_data_in_eager_boxing * refine * refine * refine * backup code * THREAD_LOCAL_CACHED * Delete thread_local_cache.h * bugfix: DeviceId4ParallelId -> MachineId4ParallelId * optimize * minor fix * LazyInterpreterApplyImplForParallelCastOpExpr * rm eager constraint * c2c interp ctx with parallel info * multi client collective boxing * test_to_consistent * support to_consistent grad_sbp * AsConsistentTensor * pass bwd test * add multi graph test * add ConsistentToConsistentOpExpr * LazyConsistentToConsistent * interpret ConsistentToConsistentOpExpr * update test * rm useless code * auto format by CI * fix conflict * mod comment * add message for local_tensor.to_consistent() check and consistent_tensor.to_local() check in lazy * address review * fix conflict * rm check which limit placement changing * auto format by CI * fix nd_sbp * auto format by CI * refactor to.py * ConsistentToConsistentOpExpr catch free tensor * fix copy op's sbp inferring * refactor empty infer sbp * refactor constant infer sbp * mod coco reader sbp inferring * fix GetSbpFn * fix consistent_to * fix (#5857) Co-authored-by: Nleaves-zwx <kunta0932@gmail.com> * modify comments * add test_to_placement case * clear code * unready test * refactor with InferNdSbp4SrcOp * rm out-dated comment * tidy code * SBP str -> cfg::SbpParallel Co-authored-by: clackhan <han_binbin@163.com> Co-authored-by: Ntsai <jackalcooper@gmail.com> Co-authored-by: NXinqi Li <lixinqi0703106@163.com> Co-authored-by: NLi Xinqi <lixinqi2010@gmail.com> Co-authored-by: Noneflow-ci-bot <ci-bot@oneflow.org> Co-authored-by: Nhjchen2 <chenhoujiangcug@gmail.com> Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Co-authored-by: NLiang Depeng <liangdepeng@gmail.com>
-
由 Li Xinqi 提交于
* GetBroadcastGroup * fix comment typo. * broadcast shape and dtype * 1) rm THREAD_LOCAL_CACHED; 2) fix bugs in ThreadLocal * fix wrong use of LocalRank * revert several code from master * fix compiler complain * merge master Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>
-