weixin_44190201的博客

cpu gpu

def x1, x2, target = sample_pair(label, emb_vec)label: [1,1,2,3,4…,1] #64emb_vec: word embedding 或者 hidden feature 等 看是要feature learning还是embedding learning #tensor [64,300]代码内部参数：r_idx: [[0,0,…0], [1,1…,1],…[63,63…63]] #[64,64]c_idx=[[0,1,2…63],…