InvGrad

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	√
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	√
Atlas inference products	√
Atlas training products	√

Function Description

• Operator function: computes the backward gradient of the Inv (reciprocal) operator, corresponding to the gradient computation of the forward y = 1 / original_x.

• Operator formula:

  $$y_i=-1\cdot x_i\cdot x_i\cdot gra{d}_i$$

  where:

  • $x_i$ is the output of the forward Inv operator (that is, $1/{\mathrm{o}\mathrm{r}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{l}\_\mathrm{x}}_i$);
  • $gra{d}_i$ is the gradient passed from upstream;
  • $y_i$ is the gradient of the original input ${\mathrm{o}\mathrm{r}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{l}\_\mathrm{x}}_i$.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x	Input	Input tensor.	FLOAT16, FLOAT32, BFLOAT16, INT32, INT8	ND
grad	Input	Corresponding input gradient.	FLOAT16, FLOAT32, BFLOAT16, INT32, INT8	ND
y	Output	Derivative gradient of the input tensor x	FLOAT16, FLOAT32, BFLOAT16, INT32, INT8	ND

Constraint Description

None

Invocation Description

Invocation Method	Sample Code	Description
Graph-mode invocation	test_geir_inv_grad	Invokes the inv_grad operator by building a graph through the operator IR.