Skip to content

Commit

Permalink
Merge pull request datawhalechina#145 from datawhalechina/2024-01
Browse files Browse the repository at this point in the history
2024年1月更新
  • Loading branch information
logan-zou authored Mar 16, 2024
2 parents faebf32 + 5552f43 commit e3cea9b
Show file tree
Hide file tree
Showing 60 changed files with 22,097 additions and 0 deletions.
8 changes: 8 additions & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -136,3 +136,11 @@ dmypy.json
.pyre/

.idea

content/Advanced Retrieval for AI with Chroma/2. 基于嵌入的检索技术概述 Overview of embeddings-based retrieval.ipynb
content/Advanced Retrieval for AI with Chroma/L3-student.ipynb
content/Advanced Retrieval for AI with Chroma/L5-Cross-encoder re-ranking.ipynb
content/Advanced Retrieval for AI with Chroma/L4-Query expansion.ipynb
content/Advanced Retrieval for AI with Chroma/L6-Embedding Adaptors.ipynb
content/Advanced Retrieval for AI with Chroma/L1-student.ipynb
content/Advanced Retrieval for AI with Chroma/images/基于Chroma的高级检索技术.xmind
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
# 基于Chroma的高级检索技术

## 简介

欢迎来到《基于Chroma的高级检索技术》课程👏🏻👏🏻

本课程由 Anton Troynikov(Chroma联合创始人)和 Deeplearning.ai 联合开发,旨在介绍基于`chroma`的高级检索技术,提升检索结果的准确性。

RAG(Retrieval Augmented Generation),又称检索增强生成,通过检索相关文本,为大语言模型(LLM)提供参考资料,从而提高LLM在特定任务上的回答准确性和可靠性。

这种方法的问题在于,嵌入模型可能会找到与查询相似主题的文本,但并不包含我们想要的答案,对LLM的回答形成干扰。为解决这个问题,我们可以采用查询扩展(Query Expansion)的方法。方式一:通过不同方式重新措辞或重写初始查询,将其扩展为多个查询;方式二:猜测或给定一个参考答案,以查看我们的文本集合中是否有更像答案的内容,而不仅仅是和查询主题有关但不含答案的干扰项。

当然,查询拓展只是其中一个方法,学完这门课程你将了解更多增强检索效果的方法,构建出更加强大的AI系统。


## 课程内容

在本课程中,您将学习到一系列知识和技术:
- 基于embedding的检索技术概述
- 检索陷阱——为何简单向量检索有时会失效
- 查询拓展
- 运用交叉编码器重新排序
- Embedding Adaptors
- 其他技术

## 致谢

最后特别感谢 Chroma 团队的 Jeff Huber, Hammad Bashir, Liquan Pei, Ben Eggers 以及 DeepLearning.AI 团队的 Geoff Ladwig 和 Esmael Gargari 对本课程做出的贡献。

Large diffs are not rendered by default.

Large diffs are not rendered by default.

Large diffs are not rendered by default.

Large diffs are not rendered by default.

Loading

0 comments on commit e3cea9b

Please sign in to comment.