Grad (#58116)

* tmp * [PIR]Migrate maximum into pir * Polish code * add ir_grad of static_gradient * add test * modify bug * modify * add mean fill_constant test * modify cpu int32 test * get_shape_tensor * delete * add default place * modify grad --------- Co-authored-by: 0x45f <[email protected]>
PaddlePaddle · Oct 17, 2023 · fb0f9c5 · fb0f9c5
1 parent 7f3b4e5
commit fb0f9c5
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/python/paddle/autograd/ir_backward.py b/python/paddle/autograd/ir_backward.py
@@ -364,8 +364,8 @@ def make_output_grad(op):
         for i, value in enumerate(op.results()):
             if (
                 value in state.value_to_valuegrad
-                and len(state.value_to_valuegrad[value])
-            ) > 1:
+                and len(state.value_to_valuegrad[value]) > 1
+            ):
                 # one value is input of more than one fwd_op,
                 # so more than one bwd_op create input_grad,
                 # need add sum op to accumulate gradient