langchain-tutorials

Some rough ideas on better Q/A using LLM+KG:

A chunk should represents a discourse pretty well. (Current RAG systems create some sort of overlapping chunks, without much idea why it's doing so...). Also, the length of this chunks effect the answer to a query, so, we should have summary of these chunks, instead their full content (incase context limit exceeds).

So we can definitely rank the chunks (can be a composite ranking: takes: into account tf+idf, semantic similarity, NE distribution etc), provided a given question.
Then we can also find the one-hop close members of those chunks, to help answer the question better. (Need to think about this ... )

Name		Name	Last commit message	Last commit date
Latest commit History 13 Commits
2024		2024
Data/Wiki-texts/Cranmer		Data/Wiki-texts/Cranmer
LCEL		LCEL
LC_Basics		LC_Basics
LLMs_APIs		LLMs_APIs
RAG		RAG
agents		agents
embeddings		embeddings
ollama		ollama
openai		openai
paper_related		paper_related
specific_llms		specific_llms
summarization		summarization
tools		tools
README.md		README.md
paths.py		paths.py