Be aware, overflowing tokens are not returned for the setting you have chosen

原文提醒如下:

Be aware, overflowing tokens are not returned for the setting you have chosen, i.e. sequence pairs with the 'longest_first' truncation strategy. So the returned list will always be empty even if some tokens have been removed.

出现场景:

encode_tokens = tokenizer.encode_plus(
                            text="1 1 1 1 1", 
                            text_pair="2 2 2",
                            padding='max_length', 
                            max_length=10, 
                            truncation=True
                )

上面的两个文本段含有8个单词(5个1,3个2),加上1个CLS、2个SEP,一共切分出11个token,大于最大长度10,会被截断,所以提醒你。

解决方案:

        1、你把上面的例子改成7个单词,不会截断,不会提醒。

        2、其他博客提到加上下面一句话:

import transformers
transformers.logging.set_verbosity_error()

        相当于改变报错级别。

mark一下,如有错误欢迎指正。

你可能感兴趣的:(transformer,transformer)