bert根据语境获得词向量,为什么词嵌入要从前面补充0

importtorchfromtorchimportnfromtransformersimportbertmodel，BertConfig， BertTokenizer#预培训模型的存储位置pre trained _ path=' Bert-base-Chinese/' config=Bert config.from _ pre trained (pre trained _ ) pre trained tokenizer=Bert tokenizer.from _ pre trained (pre trained ) ained_prom ) pre trained config=config (batch=tokenizer.# encode_plus (这是一个晴朗的早晨) ) encode _ plus是词典input _ ids=token _ type _ ids=torch.tensor ([batch] [ ' token _ type _ ids ' ] ) attention_mask=torch.tensor ) [ batch