未验证 提交 fab4e6d0 编写于 作者: Z Zhong Hui 提交者: GitHub

add abs support double grad

add abs support double grad for the api 2.0
上级 36ed83d2
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/update_ps_heter github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chenwhql/saveload/remove_save_load_config github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/iclementine/rnn_fix github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/mapingshuo/doc_2.0 github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0925 github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wzzju/fix_err_info github/fork/xiemoyuan/op_error_message github/fork/yaoxuefeng6/fix_doc github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/is_compile_with_cuda github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhiqiu/dev/refine_initializer improve_sccache incubate/infrt inplace_addto make_flag_adding_easier move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 pten_tensor_refactor release/2.0 release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment rocm_dev_0217 support_weight_transpose test_benchmark_ci test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0
无相关合并请求
......@@ -763,10 +763,28 @@ class ActivationOpDoubleGrad2 : public framework::OperatorWithKernel {
}
};
//
// AbsGrad: dx=dy if x >=0 else -dy
// AbsDoubleGrad: ddy = ddx if x >=0 else -ddx
template <typename T>
class AbsDoubleGradMaker : public ::paddle::framework::SingleGradOpMaker<T> {
public:
using ::paddle::framework::SingleGradOpMaker<T>::SingleGradOpMaker;
protected:
void Apply(GradOpPtr<T> op) const override {
op->SetType("abs_grad_grad");
// input1: x
op->SetInput("X", this->Input("X"));
// input2: ddx
op->SetInput("DDX", this->OutputGrad(framework::GradVarName("X")));
op->SetAttrMap(this->Attrs());
// output: ddy
op->SetOutput("DDOut", this->InputGrad(framework::GradVarName("Out")));
}
};
// ReluGrad: dx = dy if y >= 0 else 0
// ReluGradGrad: ddy = ddx if y >= 0 else 0
//
template <typename T>
class ReluDoubleGradMaker : public ::paddle::framework::SingleGradOpMaker<T> {
public:
......@@ -1214,7 +1232,13 @@ REGISTER_OPERATOR(
std::conditional<ops::CanInplaceAct<ops::AbsGradFunctor<float>>(),
ops::ActFwdInplaceInferer, void>::type);
REGISTER_OPERATOR(abs_grad, ops::ActivationOpGrad,
ops::ActivationGradOpInplaceInferer);
ops::ActivationGradOpInplaceInferer,
ops::AbsDoubleGradMaker<paddle::framework::OpDesc>,
ops::AbsDoubleGradMaker<paddle::imperative::OpBase>);
REGISTER_OPERATOR(
abs_grad_grad,
ops::ActivationOpDoubleGrad<ops::AbsGradGradFunctor<float>::FwdDeps()>,
ops::ActivationDoubleGradOpInplaceInferer);
REGISTER_OP_CPU_KERNEL(abs,
ops::ActivationKernel<paddle::platform::CPUDeviceContext,
......@@ -1234,6 +1258,18 @@ REGISTER_OP_CPU_KERNEL(
ops::AbsGradFunctor<int>>,
ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
ops::AbsGradFunctor<int64_t>>);
REGISTER_OP_CPU_KERNEL(
abs_grad_grad,
ops::ActivationDoubleGradKernel<plat::CPUDeviceContext,
ops::AbsGradGradFunctor<float>>,
ops::ActivationDoubleGradKernel<plat::CPUDeviceContext,
ops::AbsGradGradFunctor<double>>,
ops::ActivationDoubleGradKernel<plat::CPUDeviceContext,
ops::AbsGradGradFunctor<plat::float16>>,
ops::ActivationDoubleGradKernel<plat::CPUDeviceContext,
ops::AbsGradGradFunctor<int>>,
ops::ActivationDoubleGradKernel<plat::CPUDeviceContext,
ops::AbsGradGradFunctor<int64_t>>);
/* ========================================================================== */
/* ========================== register checkpoint ===========================*/
......
......@@ -160,7 +160,7 @@ REGISTER_OP_CUDA_KERNEL(
ops::ExpGradFunctor<plat::float16>>);
/* ========================================================================== */
/* ========================== exp register ============================ */
/* ========================== abs register ============================ */
REGISTER_OP_CUDA_KERNEL(
abs, ops::ActivationKernel<plat::CUDADeviceContext, ops::AbsFunctor<float>>,
......@@ -180,4 +180,16 @@ REGISTER_OP_CUDA_KERNEL(
ops::AbsGradFunctor<int64_t>>,
ops::ActivationGradKernel<plat::CUDADeviceContext,
ops::AbsGradFunctor<plat::float16>>);
REGISTER_OP_CUDA_KERNEL(
abs_grad_grad,
ops::ActivationDoubleGradKernel<paddle::platform::CUDADeviceContext,
ops::AbsGradGradFunctor<float>>,
ops::ActivationDoubleGradKernel<paddle::platform::CUDADeviceContext,
ops::AbsGradGradFunctor<double>>,
ops::ActivationDoubleGradKernel<plat::CUDADeviceContext,
ops::AbsGradGradFunctor<plat::float16>>,
ops::ActivationDoubleGradKernel<paddle::platform::CUDADeviceContext,
ops::AbsGradGradFunctor<int>>,
ops::ActivationDoubleGradKernel<paddle::platform::CUDADeviceContext,
ops::AbsGradGradFunctor<int64_t>>);
/* ========================================================================== */
......@@ -1430,6 +1430,27 @@ class ActivationDoubleGradKernel
}
};
template <typename T>
struct AbsGradGradFunctor : public BaseActivationFunctor<T> {
template <typename Device>
void operator()(const Device& dev, const framework::Tensor* X,
const framework::Tensor* Out, const framework::Tensor* ddX,
framework::Tensor* ddOut, framework::Tensor* dOut,
framework::Tensor* dX) const {
auto* d = dev.eigen_device();
auto ddx = framework::EigenVector<T>::Flatten(
GET_DATA_SAFELY(ddX, "Input", "DDX", "AbsGradGrad"));
auto x = framework::EigenVector<T>::Flatten(
GET_DATA_SAFELY(X, "Input", "X", "AbsGradGrad"));
if (ddOut) {
auto ddout = framework::EigenVector<T>::Flatten(
GET_DATA_SAFELY(ddOut, "Output", "DDOut", "AbsGradGrad"));
ddout.device(*d) = ddx * x.sign();
}
}
static constexpr ActBwdOpFwdDeps FwdDeps() { return kDepX; }
};
template <typename T>
struct ReluGradGradFunctor : public BaseActivationFunctor<T> {
template <typename Device>
......
......@@ -147,5 +147,29 @@ class TestSquareDoubleGradCheck(unittest.TestCase):
self.func(p)
class TestAbsDoubleGradCheck(unittest.TestCase):
@prog_scope()
def func(self, place):
# the shape of input variable should be clearly specified, not inlcude -1.
shape = [2, 3, 7, 9]
eps = 1e-6
dtype = np.float64
x = layers.data('x', shape, False, dtype)
x.persistable = True
y = layers.abs(x)
x_arr = np.random.uniform(-1, 1, shape).astype(dtype)
gradient_checker.double_grad_check(
[x], y, x_init=x_arr, place=place, eps=eps)
def test_grad(self):
places = [fluid.CPUPlace()]
if core.is_compiled_with_cuda():
places.append(fluid.CUDAPlace(0))
for p in places:
self.func(p)
if __name__ == "__main__":
unittest.main()
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册
反馈
建议
客服 返回
顶部