LogAddExp
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 logarithm of the sum of exponentials of two input tensors element-wise. Under the default attributes, the base is e.
- Computation formula:
y=ln(ex1+ex2)y = \ln\left(e^{x_1} + e^{x_2}\right)
Parameter Description
| Parameter Name | Input/Output/Attribute | Description | Data Type | Data Format |
|---|---|---|---|---|
| x1 | Input | Input tensor x_1 in the formula | FLOAT16, BFLOAT16, FLOAT32 | ND |
| x2 | Input | Input tensor x2 in the formula | Same as x1 | ND |
| base | Optional attribute | Base of the exponential and logarithm operations, with a default value of -1.0, indicating the base e | FLOAT32 | - |
| scale | Optional attribute | Scaling coefficient of -abs(x1-x2) in the exponential term, with a default value of 1.0 | FLOAT32 | - |
| shift | Optional attribute | Offset of the exponential term, with a default value of 0.0 | FLOAT32 | - |
| y | Output | Output tensor y in the formula | Same as x1 | ND |
Constraint Description
- base must be -1.0 or a value greater than 0.
- The data types of x1, x2, and y must be the same.
- The shapes of x1 and x2 must satisfy the broadcast relationship, and the shape of y is the shape after broadcasting.
- Atlas inference products and Atlas training products do not support the BFLOAT16 data type.
Invocation Description
| Invocation Method | Invocation Sample | Description |
|---|---|---|
| aclnn invocation | test_aclnn_log_add_exp | Invokes the LogAddExp operator through the aclnnLogAddExp API |
| aclnn invocation | test_aclnn_log_add_exp2 | Invokes the LogAddExp operator through the aclnnLogAddExp2 API |