HardSigmoidBackward¶
General¶
HardSigmoidBackward operation computes gradient for HardSigmoid. The formula is defined as follows:
\[\begin{split}diff\_src = \begin{cases} diff\_dst \cdot \alpha & \text{if}\ 0 < \alpha src + \beta < 1 \\ 0 & \text{otherwise}\ \end{cases}\end{split}\]
Operation attributes¶
Attribute Name |
Description |
Value Type |
Supported Values |
Required or Optional |
|---|---|---|---|---|
\(\alpha\) in the formula. |
f32 |
Arbitrary f32 value. |
Required |
|
\(\beta\) in the formula. |
f32 |
Arbitrary f32 value. |
Required |
Execution arguments¶
The inputs and outputs must be provided according to the index order shown below when constructing an operation.
Inputs¶
Index |
Argument Name |
Required or Optional |
|---|---|---|
0 |
|
Required |
1 |
|
Required |
Outputs¶
Index |
Argument Name |
Required or Optional |
|---|---|---|
0 |
|
Required |
Supported data types¶
HardSigmoidBackward operation supports the following data type combinations.
Src |
Diff_dst |
Diff_src |
|---|---|---|
f32 |
f32 |
f32 |
f16 |
f16 |
f16 |
bf16 |
bf16 |
bf16 |