Deep dive into RAG Chunking Strategies : CharacterText Splitter to Semantic Chunking

- Why do we need chunking in RAG
- Different chunking strategies and Pros and Cons
- CharacterText Splitter
- RecursiveCharacter Text Splitter
- TokenText Splitter
- based embedding max tokens
- LLM Context length
- Semantic Chunking
Notebook: github.com/ari...

[@awakenwithoutcoffee]

I appreciate this excellent breakdown Artira . The semantic chunker is something that could be a real breaktrough for technical documentations. One of the problems we keep facing is that "summarization" is often leaving out too many details (e.g.: techincal lists that are incomplete, instructions that should NOT be summarized are still summarized etc.).😶

[@AritraSen]

Hey, glad you liked it :)
One of the thing you can try out is giving instructions in the prompt and ask the model to think in a step by step fashion with those instructions.

[@oguzhanylmaz4586]

Hello, I want to make a project like this: fast api or flask api + reactjs + (open source llm (Llama2, 3, mistral etc.). But I couldn't find any project that I can reference. Can you make such a project?
I don't want to use openai. The project must be able to run without internet.

[@AritraSen]

We already have a playlist - RAG LLM App with FastAPI and Gradio: th-cam.com/play/PLOrU905yPYXIqQLY6ulQqB8e414-DFuyd.html

[@awakenwithoutcoffee]

@@AritraSen I think OP is looking to integrate a React front-end using an API. Gradio seems good for a demo but React is allot more customizable.

[@AritraSen]

@@awakenwithoutcoffee hey sorry React is not my skill set...

[@awakenwithoutcoffee]

@@AritraSen totally fine brother, keep on creating !