AscendAntiQuant

Description

Performs requantization on the input x.

Input

x: tensor of type int8. The data format is ND.

Attribute

• offset: float, specifying the offset.
• scale: float, specifying the scale factor.
• (Optional) sqrt_mode: bool, whether to perform square root calculation on the scale. Defaults to false.

Output

y: tensor of type float16. The data format is ND.

Constraints

• For Atlas inference products , the data format of the input scale, input offset, and output y cannot be bfloat16.
• For Atlas inference products , the data format of the input x cannot be int4 or int32.

ONNX Opset Support

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