← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6623 · 生成模型:GPT-6

训练分词器与调用分词器分开

理解等级:中级

训练阶段由语料学习词表或切分规则;编码阶段按已确定配置把文本转换为ID。预训练模型通常需要其匹配的分词配置,不能只在推理时重训或替换tokenizer就假定权重仍正确理解输入。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      A tokenizer has two responsibilities: In the tokenizer pre-training stage, the tokenizer is run over a body of text to generate a vocabulary. While processing input during both model training and inference, free-form raw text is run through the tokenizer algorithm to break the text into sequences of valid tokens. Figure 3-2 depicts the roles played by a tokenizer. Figure 3-2. Tokenizer workflow When we

    参考来源

    - 位置:《Designing Large Language Model Applications》第03章《词表与分词》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch03.html) - 结论依据:摘录中关于训练分词器与调用分词器分开的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:A tokenizer has two responsibilities: In the tokenizer pre-training stage, the tokenizer is run over a body of text to generate a vocabulary. While processing input during both model training and inference, free-form raw text is run through the tokenizer algorithm to break the text into sequences of valid tokens. Figure 3-2 depicts the roles played by a tokenizer. Figure 3-2. Tokenizer workflow When we