search

labring / FastGPT

FastGPT is a knowledge-based platform built on the LLMs, offers a comprehensive suite of out-of-the-box capabilities such as data processing, RAG retrieval, and visual AI workflow orchestration, letting you easily develop and deploy complex question-answering systems without the need for extensive setup or configuration.
https://tryfastgpt.ai
Other
18.18k stars 4.82k forks source link

Can Microsoft's Graph RAG be added to the new version? #1969

Open jeffreychen567 opened 4 months ago

jeffreychen567 commented 4 months ago

The current RAG still has numerous deficiencies. Microsoft's Graph RAG could substantially improve RAG capabilities, and I hope it will be incorporated into the next version.

anfernees-iphone commented 4 months ago

+1

Essence9999 commented 4 months ago

+1

Tining123 commented 4 months ago

+1

dockersky commented 4 months ago

+1

Code-Bai commented 4 months ago

+1

sosmygirl commented 4 months ago

+1

zhouxihong1 commented 4 months ago

+1

dongfangzan commented 4 months ago

+1

c121914yu commented 4 months ago

目前结论，太贵了，≈概念，没太多实际场景价值。单个问题成本5块左右，比起普通方案贵了20倍（不如直接把普通方案增加20倍 token，效果也不差，top10变成top200）

anfernees-iphone commented 3 months ago

目前结论，太贵了，≈概念，没太多实际场景价值。单个问题成本5块左右，比起普通方案贵了20倍（不如直接把普通方案增加20倍 token，效果也不差，top10变成top200）

召回的内容多了，也考验LLM的归纳整理能力

dfrrk commented 3 months ago

强烈希望Graph RAG它能够被纳入到下一个版本中更新！！

GrayXu commented 3 months ago

目前结论，太贵了，≈概念，没太多实际场景价值。单个问题成本5块左右，比起普通方案贵了20倍（不如直接把普通方案增加20倍 token，效果也不差，top10变成top200）

还是不太一样吧，top200的一个问题也是拿了太多没用信息，ref拉了一大堆，人工也检索不了这么多ref。
graphrag的kg可以做更细粒度的知识。而且embedding也会有幻觉。

Mgrsc commented 3 months ago

强烈希望Graph RAG它能够被纳入到下一个版本中更新！！

buctllx commented 2 months ago

+1

dvlin-dev commented 2 months ago

+1