Skip to content

V3版本 规划 #29

@lloydzhou

Description

@lloydzhou
  1. 使用inference API GPTS 支持 向量搜索
    a. 通过pipeline自动处理向量化过程(支持huggingface api 以及openai api)
    b. 通过query_vector_builder在knn查询阶段也使用inference进行向量化
  2. 使用huggingface/text-embeddings-inference 作为api移除之前内置pytouch做向量化的过程,这里可以提升之前知识库做向量化的性能
  3. 利用es 支持 inner hits的特性 改进存储结构
  4. 文档拆分https://github.com/Filimoa/open-parse
    a. 使用这个开源的open parse项目。对pdf支持挺好。
    b. 另外就是这个项目readme提到的 google document ai,以及aws的相关api,还有一个公司的产品(这些都是付费的 $10 / 1000page)
    • Typically priced at ≈ $10 / 1k pages. See here, here and here.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions