weighted loss for different class 为不同数据的loss分配不同的权重

[Li-Jicheng]

Hi, I'm wondering if there's a way to assign different weights on different data samples.

For example, the input data contains sample for task1, task2, task3, .... 。And samples from different tasks can be easily distinguished, either by sample index or prompt.

How can I assign more penalty on task2 for misclassification?

Thanks