fixed arg_max

mindspore/ops/_op_impl/tbe/arg_max.py

@@ -23,8 +23,8 @@ arg_max_op_info = TBERegOp("Argmax") \
     .compute_cost(10) \
     .kernel_name("arg_max_d") \
     .partial_flag(True) \
-    .attr("dimension", "required", "int", "all") \
-    .attr("dtype", "optional", "type", "all") \
+    .attr("axis", "required", "int", "all") \
+    .attr("output_dtype", "optional", "type", "all") \
     .input(0, "x", False, "required", "all") \
     .output(0, "y", False, "required", "all") \
     .dtype_format(DataType.F16_Default, DataType.I32_Default) \

mindspore/ops/operations/array_ops.py

@@ -951,8 +951,8 @@ class Argmax(PrimitiveWithInfer):
 
     Args:
         axis (int): Axis on which Argmax operation applies. Default: -1.
-        output_type (:class:`mindspore.dtype`): An optional data type of `mindspore.dtype.int32` and
-            `mindspore.dtype.int64`. Default: `mindspore.dtype.int64`.
+        output_type (:class:`mindspore.dtype`): An optional data type of `mindspore.dtype.int32`.
+            Default: `mindspore.dtype.int32`.
 
     Inputs:
         - **input_x** (Tensor) - Input tensor.
@@ -961,12 +961,12 @@ class Argmax(PrimitiveWithInfer):
         Tensor, indices of the max value of input tensor across the axis.
 
     Examples:
-        >>> input_x = Tensor(np.array([2.0, 3.1, 1.2]))
+        >>> input_x = Tensor(np.array([2.0, 3.1, 1.2]), mindspore.float32)
         >>> index = P.Argmax(output_type=mindspore.int32)(input_x)
     """
 
     @prim_attr_register
-    def __init__(self, axis=-1, output_type=mstype.int64):
+    def __init__(self, axis=-1, output_type=mstype.int32):
         """init Argmax"""
         self.init_prim_io_names(inputs=['x'], outputs=['output'])
         validator.check_value_type("axis", axis, [int], self.name)