Transformer 最早在 Attention Is All You Need 中以机器翻译架构的形式提出。理解它的一个起点是：每个位置怎样从其他位置取回当前需要的信息？
这篇旧稿于 2026 年 9 月补全。保留原始链接与项目背景，修正了任务分类、预测评估等表述。示例代码用于理解模型结构，不代表训练结果。
Transformer 最早在 Attention Is All You Need 中以机器翻译架构的形式提出。理解它的一个起点是：每个位置怎样从其他位置取回当前需要的信息？
这篇旧稿于 2026 年 9 月补全。保留原始链接与项目背景，修正了任务分类、预测评估等表述。示例代码用于理解模型结构，不代表训练结果。
自然语言处理（NLP）研究怎样让计算机处理人类语言。这里不从复杂模型开始，而是沿着“任务 → 数据 → 表示 → 模型 → 评估”走一遍，把第一条可运行的流程搭起来。
原文发表于 2022 年；本次正文于 2026 年 9 月补全。下方数据为自拟教学样本，不用于证明任何模型的真实性能。
The original Attention Is All You Need introduced Transformer as an encoder–decoder architecture for machine translation. A useful starting question is: how does each position retrieve the information it needs from other positions?
This note retains its original publication date and URL. It was completed and revised in September 2026. The code below demonstrates tensor flow, not trained forecasting performance.
安装 Python 不只是让终端能输出一个版本号。真正需要确认的是：项目使用哪个解释器，依赖安装在哪里，以及换一台电脑后怎样重建环境。
这篇笔记保留原始发表日期，正文于 2026 年 9 月补全。命令采用 Python 3 与内置
venv；安装界面和可用版本请以官方文档为准。
总而言之，虽然博客的内容可能各种各样，但是整体上是我一段时间的学习心得和经验总结，有没有可能有生活上呢，应该是有的。大部分的博客我都尽可能撰写中英双语，当然如果我后期日语学的也比较好就可能变成三语，谁知道呢。
至于博客我都打上tag，这样也方便大家进行查阅。
Natural language processing is about working with human language computationally. This introduction follows a practical path: define the task, prepare the data, represent the text, train a baseline, and evaluate it.
Originally published in 2022; completed and revised in September 2026. The examples below use a tiny, invented dataset for teaching, not for benchmarking.
This is my blog. I think that I will write a lot of things in the future. It mainly about some my experience in computer learning. It may have many fields because I like share my experience in all fields.
All in all, this blog’s content will be various. But it mostly will be the experience for my in a period of time. Is there any about my life? Probably have. Most my blog will write both in English and Chinese. If I learn Japanese well in the future, maybe I will write in three different languages. Who knows.
I will add tag in every post in order to help you look up conveniently.