FilesLast commitLast update
4 months ago
3 months ago
25 days ago
9 days ago
9 days ago
10 days ago
9 days ago
1 month ago
18 days ago
README

ApplyGradientDescent

English | 简体中文

Applicable Products

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

Function

  • Operator function: Updates model parameters during gradient descent training.

  • Formula:

    var=var−alpha×deltavar = var - alpha \times delta

    Here, var denotes the model parameters to be updated, alpha denotes the learning rate, and delta denotes the gradient of the parameters.

Parameters

Parameter name Input/Output/Attribute Description Data type Data format
var Input The model parameters to be updated, var in the formula. FLOAT, FLOAT16, BFLOAT16 ND
alpha Input The learning rate, which controls the step size of parameter updates, alpha in the formula. FLOAT, FLOAT16, BFLOAT16 ND
delta Input The gradient value of parameter var, delta in the formula. FLOAT, FLOAT16, BFLOAT16 ND
var Output The updated model parameters, var in the formula. FLOAT, FLOAT16, BFLOAT16 ND

Constraints

None

Invocation Description

Invocation method Invocation sample Description
Graph-mode invocation test_geir_apply_gradient_descent Calls the ApplyGradientDescent operator by building a graph through the operator IR.