AscendAntiQuant

Description

Perform dequantization on input x.

Input

One input:

x: Input tensor. The data type supports int8, and the data format supports ND.

Output

One output:

y: tensor of type float16.

Attribute

offset: float, specifying the offset.

scale: float, specifying the scaling ratio.

sqrt_mode: bool, specifying whether to perform square root calculation on the scale. The default value is false. This attribute is optional.

Constraints

For the Atlas inference series products, the input parameter scale, input parameter offset, and output parameter y do not support the bfloat16 data format.

For the Atlas inference series products, the input parameter x does not support the int4 or int32 data format.

ONNX Opset Support

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