https://github.com/kerthcet/k8s-specific-knowledge-base

A QA system based on k8s-specific knowledge build on ChatGLM2-6B, serving by Ray.
https://github.com/kerthcet/k8s-specific-knowledge-base

Last synced: 8 days ago
JSON representation

A QA system based on k8s-specific knowledge build on ChatGLM2-6B, serving by Ray.

• Host: GitHub
• URL: https://github.com/kerthcet/k8s-specific-knowledge-base
• Owner: kerthcet
• Created: 2023-07-26T09:56:21.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-09-14T11:48:48.000Z (about 1 year ago)
• Last Synced: 2023-09-15T17:56:37.496Z (about 1 year ago)
• Language: Python
• Size: 44 MB
• Stars: 7
• Watchers: 2
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# k8s-specific-knowledge-base

This is a Question-Answering system based on k8s-specific knowledge build on ChatGLM2-6B, serving by Ray.

It's not ready for production, but we'll target for this.

## Architecture

![architecture](./images/architecture.png)

## 🤔 What we used ?

- [LangChain](https://github.com/langchain-ai/langchain)  to build LLM applications.

- [Ray](https://github.com/ray-project/ray) for accelerating and serving.

- [ChatGLM2-6B](https://github.com/THUDM/ChatGLM2-6B) as a base model.

- [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh) for embedding in semantic search.

- [FAISS](https://github.com/facebookresearch/faiss) as a vector database.

## 📦 Corpus Including

- Kubernetes Website

- Kubernetes Blogs

- Kubernetes Books (Only for research usage)

## 🔖 What's next ?

- Containerization, FAISS is for single node.

- More efficient text splitting ways designed for Chinese.

- More approaches to support semantic search, e.g. key-word embeddings.

- Raw data management, like new uploading and deleting.

- Vector data persistent.

- Continuous Pre-Training.

- and so on ...

## 📚 Reference

- [langchain-ray](https://github.com/ray-project/langchain-ray)