2023-03-24干活小计

昨天上了一天课摸鱼，就不记录了
在dataset中写collate_fn，对batch data解压操作 用zip(*)比较便捷

def data_process(self, batch_data):
    batch_txt = []
    batch_label = []
    batch_len = []
    for data in batch_data:
        batch_txt.append(data[0])
        batch_label.append(data[1])
        batch_len.append(data[2])
    #等价于        a, b, c = zip(*batch_data)


又被buding爸爸装到了，bert用0.01学习率直接损失不动了，用0.00001甚至更小