Hey all looking into a couple frameworks

BBeff Jezos

Hey all, looking into a couple frameworks including LlamaIndex. I’m curious if I’m maybes over thinking it.

Ultimately, here is my goal

I want a train datasets based on user input

There are 3 main types of dataset
-industry knowledge base
-transcriptions
-website scrap crawls

I figured you can use the direct OpenAI to do various embedding.

Then when I provide GPT4 with a prompt, it will look up a specific dataset row or even the whole dataset.

Anyone here think a framework like LlamaIndex will be efficient or is it an overkill?

10 comments

LLogan M

This sounds exactly like what llamaindex is made for 🙂

BBrad

I've built a couple PoC's for each of those use cases. You can go as high or deep as you want with LlamaIndex for these.

BBeff Jezos

@Logan M haha that’s what I figured. I looked into txtai too but it doesn’t really integrate with OpenAI

BBeff Jezos

Are there some low code platforms to launch concepts fast ?

BBeff Jezos

@Brad do you mind shoewcasing

BBrad

They're work product, can't share, sadly.

BBrad

If you have a look at the docs you can get a PoC going really quick with very little code. The docs are great.

To get a really usable app, especially in production, it's like anything else. Have to invest the time learning and developing.

BBeff Jezos

@Brad you prefer llama over langchain?

BBrad

Different tools, with overlap. You can use them both together if you like. LlamaIndex has the benefit of tight focus on RAG from Jerry, Logan, et al. And the docs explaining RAG concepts are great.

If you're building say, a multi-agent tool using system, you might use LlamaIndex over data and LangChain for other agents in the system.

BBeff Jezos

@Logan M hey so I saw your message here that you can't use gpt-4 to finetune

https://discord.com/channels/1059199217496772688/1157778390737244239/1157795836340740217

While I understand the latest gpt model you can fine-tune is 3.5

I'd like to explain to you an example and let me know if I am on the right track.

I want to give ChatGPT a knowledge base. Let's say I wanted to give the knowledge base various books to read about Sales

Then I want to label each book say a type of sales method
Persuasive
Consultative
High-pressure

Then I want to tell GPT4 to give explain to me what the Perusaive Sales method is.

My question is how does GPT4 know what to take from the knowledge base. Would I need to give GPT4 a different output?

Example Workflow:

Create Knowledge Base
User Query Summary Index
Output of Summary Index is saved as Summary_1
Chat GPT4 Prompt below

"Please explain to me the Persuasive Sales Method by using the context below as a reference

[Summary_1]"

Add a reply

Find answers from the community

Hey all looking into a couple frameworks