codefuse-ai / ModelCache

A LLM semantic caching system aiming to enhance user experience by reducing response time via cached query-result pairs.
Other
892 stars 44 forks source link

[编程挑战季] ModelCache检索的Reranker 能力 #55

Open peng3307165 opened 1 month ago

peng3307165 commented 1 month ago
  1. 问题简述 通过向量数据库检索embedding作为初次召回,仅通过L2或者cosine计算相似度,无法保障召回的query和用户的query完全匹配,希望完善Reranker能力,使用reranker模型进行二次排序,增加Cache的命中的准确性。
  2. 期望产出 在ModelCache的 adapter_query 模块增加Reranker模块,实现二次排序能力。
  3. 能力要求 熟悉python代码,了解prompt工程。
charleschile commented 1 week ago

Please assign this to me, I will try to use bge reranker and other open-source models.

DreamCyc commented 1 week ago

Oh, I'm late. If the previous student is too busy to handle it, I can assist in completing it. Additionally, the BGE embedding model is quite effective in terms of representation capability.