ReduceMax

Description

Reduces a dimension of a tensor by calculating the maximum value of the input tensor along the specified dimension. If keepdims is set to 1, the number of dimensions of the generated tensor is the same as that of the input tensor. If keepdims is set to 0, the generated tensor has the dimension pruned.

Input

data: input tensor of type float16, float, int8, int32, int64, or double.

Output

reduced data: output tensor of the identical data type as input data.

Attribute

• axes: int list, specifying the calculation axis. Value range: [-r, r-1], where r indicates the number of dimensions of the input data.
• keepdims(default is "1"): int, indicating whether to retain the reduced dimension. The default value is 1 (reserved).

Constraints

If the input is of type float16, float, or int32, the value of axes can be empty. If the input is of other types, the value of axes must be configured.

If the value of axes is empty, the full-dimension specification is not performed on the current output. In this case, you are advised to upgrade the ONNX operator to include all axes. Alternatively, before exporting the ONNX graph from Torch, use amax in the Torch model to reduce all axes. Example formula: x.amax(dim=[0, 1, 2])

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16/v17/v18