delete remote_prefetch (#52748)

paddle/fluid/operators/hierarchical_sigmoid_op.cc

@@ -124,7 +124,6 @@ class HierarchicalSigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
     AddAttr<AttrType>("num_classes", "(int, optional), The number of classes")
         .SetDefault(2);
     // for parameter prefetch
-    AddAttr<bool>("remote_prefetch", "").SetDefault(false);
     AddAttr<int>("trainer_id", "trainer id from 0 ~ worker_num.").SetDefault(0);
     AddAttr<std::vector<int64_t>>("height_sections",
                                   "Height for each output SelectedRows.")

paddle/phi/api/yaml/legacy_backward.yaml

@@ -475,8 +475,8 @@
     func : heaviside_grad
 
 - backward_op : hsigmoid_loss_grad
-  forward : hsigmoid_loss (Tensor x, Tensor label, Tensor w, Tensor bias, Tensor path, Tensor code, int num_classes, bool remote_prefetch, bool is_sparse) -> Tensor(out), Tensor(pre_out), Tensor(w_out)
-  args : (Tensor x, Tensor w, Tensor label, Tensor path, Tensor code, Tensor bias, Tensor pre_out, Tensor out_grad, int num_classes, bool remote_prefetch, bool is_sparse)
+  forward : hsigmoid_loss (Tensor x, Tensor label, Tensor w, Tensor bias, Tensor path, Tensor code, int num_classes, bool is_sparse) -> Tensor(out), Tensor(pre_out), Tensor(w_out)
+  args : (Tensor x, Tensor w, Tensor label, Tensor path, Tensor code, Tensor bias, Tensor pre_out, Tensor out_grad, int num_classes, bool is_sparse)
   output : Tensor(x_grad), Tensor(w_grad), Tensor(bias_grad)
   infer_meta :
     func : GeneralTernaryGradInferMeta

paddle/phi/api/yaml/legacy_ops.yaml

@@ -684,7 +684,7 @@
   backward : heaviside_grad
 
 - op : hsigmoid_loss
-  args : (Tensor x, Tensor label, Tensor w, Tensor bias, Tensor path, Tensor code, int num_classes, bool remote_prefetch, bool is_sparse)
+  args : (Tensor x, Tensor label, Tensor w, Tensor bias, Tensor path, Tensor code, int num_classes, bool is_sparse)
   output : Tensor(out), Tensor(pre_out), Tensor(w_out)
   infer_meta :
     func : HSigmoidLossInferMeta

paddle/phi/infermeta/multiary.cc

@@ -1432,7 +1432,6 @@ void HSigmoidLossInferMeta(const MetaTensor& x,
                            const MetaTensor& path,
                            const MetaTensor& code,
                            int num_classes,
-                           bool remote_prefetch,
                            bool is_sparse,
                            MetaTensor* out,
                            MetaTensor* pre_out,

paddle/phi/infermeta/multiary.h

@@ -312,7 +312,6 @@ void HSigmoidLossInferMeta(const MetaTensor& x,
                            const MetaTensor& path,
                            const MetaTensor& code,
                            int num_classes,
-                           bool remote_prefetch,
                            bool is_sparse,
                            MetaTensor* out,
                            MetaTensor* pre_out,

paddle/phi/kernels/cpu/hsigmoid_loss_grad.h

@@ -35,7 +35,6 @@ void HSigmoidLossGradKernelImpl(const Context& ctx,
                                 const DenseTensor& pre_out,
                                 const DenseTensor& out_grad,
                                 int num_classes,
-                                bool remote_prefetch,
                                 bool is_sparse,
                                 DenseTensor* x_grad,
                                 DenseTensor* w_grad,

paddle/phi/kernels/cpu/hsigmoid_loss_grad_kernel.cc

@@ -31,7 +31,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                             const DenseTensor& pre_out,
                             const DenseTensor& out_grad,
                             int num_classes,
-                            bool remote_prefetch,
                             bool is_sparse,
                             DenseTensor* x_grad,
                             DenseTensor* w_grad,
@@ -46,7 +45,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                                 pre_out,
                                 out_grad,
                                 num_classes,
-                                remote_prefetch,
                                 is_sparse,
                                 x_grad,
                                 w_grad,

paddle/phi/kernels/cpu/hsigmoid_loss_kernel.cc

@@ -34,7 +34,6 @@ void HSigmoidLossKernel(const Context& ctx,
                         const paddle::optional<DenseTensor>& path,
                         const paddle::optional<DenseTensor>& code,
                         int num_classes,
-                        bool remote_prefetch,
                         bool is_sparse,
                         DenseTensor* out,
                         DenseTensor* pre_out,

paddle/phi/kernels/hsigmoid_loss_grad_kernel.h

@@ -29,7 +29,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                             const DenseTensor& pre_out,
                             const DenseTensor& out_grad,
                             int num_classes,
-                            bool remote_prefetch,
                             bool is_sparse,
                             DenseTensor* x_grad,
                             DenseTensor* w_grad,

paddle/phi/kernels/hsigmoid_loss_kernel.h

@@ -27,7 +27,6 @@ void HSigmoidLossKernel(const Context& ctx,
                         const paddle::optional<DenseTensor>& path,
                         const paddle::optional<DenseTensor>& code,
                         int num_classes,
-                        bool remote_prefetch,
                         bool is_sparse,
                         DenseTensor* out,
                         DenseTensor* pre_out,

paddle/phi/kernels/selected_rows/hsigmoid_loss_grad_kernel.cc

@@ -48,7 +48,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                             const DenseTensor& pre_out,
                             const DenseTensor& out_grad,
                             int num_classes,
-                            bool remote_prefetch,
                             bool is_sparse,
                             DenseTensor* x_grad,
                             SelectedRows* w_grad,
@@ -74,7 +73,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                                      pre_out,
                                      out_grad,
                                      num_classes,
-                                     remote_prefetch,
                                      is_sparse,
                                      x_grad,
                                      w_grad_value,

paddle/phi/kernels/selected_rows/hsigmoid_loss_grad_kernel.h

@@ -31,7 +31,6 @@ void HSigmoidLossGradKernel(const Context& ctx,
                             const DenseTensor& pre_out,
                             const DenseTensor& out_grad,
                             int num_classes,
-                            bool remote_prefetch,
                             bool is_sparse,
                             DenseTensor* x_grad,
                             SelectedRows* w_grad,

paddle/phi/ops/compat/hierarchical_sigmoid_sig.cc

@@ -20,7 +20,7 @@ KernelSignature HierarchicalSigmoidOpArgumentMapping(
     const ArgumentMappingContext& ctx) {
   return KernelSignature("hsigmoid_loss",
                          {"X", "Label", "W", "Bias", "PathTable", "PathCode"},
-                         {"num_classes", "remote_prefetch", "is_sparse"},
+                         {"num_classes", "is_sparse"},
                          {"Out", "PreOut", "W_Out"});
 }
 
@@ -36,7 +36,7 @@ KernelSignature HierarchicalSigmoidGradOpArgumentMapping(
                             "Bias",
                             "PreOut",
                             "Out@GRAD"},
-                           {"num_classes", "remote_prefetch", "is_sparse"},
+                           {"num_classes", "is_sparse"},
                            {"X@GRAD", "W@GRAD", "Bias@GRAD"});
   } else if (ctx.IsSelectedRowsOutput("W@GRAD")) {
     return KernelSignature("hsigmoid_loss_grad_sr",
@@ -48,7 +48,7 @@ KernelSignature HierarchicalSigmoidGradOpArgumentMapping(
                             "Bias",
                             "PreOut",
                             "Out@GRAD"},
-                           {"num_classes", "remote_prefetch", "is_sparse"},
+                           {"num_classes", "is_sparse"},
                            {"X@GRAD", "W@GRAD", "Bias@GRAD"});
   } else {
     return KernelSignature("unregistered", {}, {}, {});

python/paddle/fluid/tests/unittests/test_hsigmoid_op.py

@@ -177,7 +177,6 @@ def python_api(
     path_code=None,
     num_classes=-1,
     is_sparse=False,
-    remote_prefetch=False,
 ):
     return paddle.nn.functional.hsigmoid_loss(
         input,

python/paddle/nn/functional/loss.py

@@ -1016,7 +1016,6 @@ def hsigmoid_loss(
         attrs = {
             "num_classes": num_classes,
             "is_sparse": is_sparse,
-            "remote_prefetch": is_sparse,
         }
 
         inputs = {