support auto generate for kldiv_loss (#51886)

PaddlePaddle · Mar 28, 2023 · cdba7e3 · cdba7e3
1 parent a2d3c33
commit cdba7e3
Show file tree

Hide file tree

Showing 7 changed files with 28 additions and 202 deletions.
diff --git a/paddle/fluid/operators/kldiv_loss_op.cc b/paddle/fluid/operators/kldiv_loss_op.cc
diff --git a/paddle/phi/api/yaml/backward.yaml b/paddle/phi/api/yaml/backward.yaml
@@ -754,6 +754,17 @@
   kernel :
     func : inverse_grad
 
+- backward_op : kldiv_loss_grad
+  forward : kldiv_loss(Tensor x, Tensor label, str reduction="mean") -> Tensor(out)
+  args : (Tensor x, Tensor label, Tensor out_grad, str reduction)
+  output : Tensor(x_grad)
+  infer_meta :
+    func : UnchangedInferMeta
+    param: [x]
+  kernel :
+    func : kldiv_loss_grad
+  no_need_buffer : x
+
 - backward_op : kron_grad
   forward : kron (Tensor x, Tensor y) -> Tensor(out)
   args : (Tensor x, Tensor y, Tensor out_grad)

diff --git a/paddle/phi/api/yaml/legacy_backward.yaml b/paddle/phi/api/yaml/legacy_backward.yaml
@@ -577,17 +577,6 @@
   optional : scale
   backward : instance_norm_double_grad
 
-- backward_op : kldiv_loss_grad
-  forward : kldiv_loss(Tensor x, Tensor label, str reduction) -> Tensor(out)
-  args : (Tensor x, Tensor label, Tensor out_grad, str reduction)
-  output : Tensor(x_grad)
-  infer_meta :
-    func : UnchangedInferMeta
-    param: [x]
-  kernel :
-    func : kldiv_loss_grad
-  no_need_buffer : x
-
 - backward_op : layer_norm_grad
   forward : layer_norm (Tensor x, Tensor scale, Tensor bias, float epsilon, int begin_norm_axis) -> Tensor(out), Tensor(mean), Tensor(variance)
   args : (Tensor x,  Tensor scale, Tensor bias, Tensor mean, Tensor variance, Tensor out_grad, float epsilon, int begin_norm_axis)

diff --git a/paddle/phi/api/yaml/legacy_ops.yaml b/paddle/phi/api/yaml/legacy_ops.yaml
@@ -825,16 +825,6 @@
   intermediate : saved_mean, saved_variance
   backward : instance_norm_grad
 
-- op : kldiv_loss
-  args : (Tensor x, Tensor label, str reduction)
-  output : Tensor(out)
-  infer_meta :
-    func : KLDivInferMeta
-  kernel :
-    func : kldiv_loss
-    data_type : x
-  backward : kldiv_loss_grad
-
 - op : layer_norm
   args : (Tensor x, Tensor scale, Tensor bias, float epsilon, int begin_norm_axis)
   output : Tensor(out), Tensor(mean), Tensor(variance)

diff --git a/paddle/phi/api/yaml/op_compat.yaml b/paddle/phi/api/yaml/op_compat.yaml
@@ -990,6 +990,13 @@
   outputs :
     out : Out
 
+- op : kldiv_loss
+  backward : kldiv_loss_grad
+  inputs :
+    {x : X, label : Target}
+  outputs :
+    out : Loss
+
 - op : kron
   backward : kron_grad
   inputs :

diff --git a/paddle/phi/api/yaml/ops.yaml b/paddle/phi/api/yaml/ops.yaml
@@ -784,6 +784,16 @@
     func : isnan {dense -> dense},
            isnan_sr {selected_rows -> selected_rows}
 
+- op : kldiv_loss
+  args : (Tensor x, Tensor label, str reduction = "mean")
+  output : Tensor(out)
+  infer_meta :
+    func : KLDivInferMeta
+  kernel :
+    func : kldiv_loss
+    data_type : x
+  backward : kldiv_loss_grad
+
 - op : kron
   args : (Tensor x, Tensor y)
   output : Tensor

diff --git a/paddle/phi/ops/compat/kldiv_loss_sig.cc b/paddle/phi/ops/compat/kldiv_loss_sig.cc