Hello

ssunwoong

Hello.
I'm working on a project about generating some sentences based on specific documents. Should I use llama_index here? I'm wondering about the difference between purpose of use of llama_index and lang_chain. Since I'm currently doing task about 'text generation', can llama_index only be used for 'Qusetion-Answering' Task?

11 comments

LLogan M

Llama Index can definitely be used for more than QA, depending on the prompt you give the system. If you have a lot of text and want to generate similar text, I think Llama Index is a good choice.

LangChain is more of a chat-frontend when you need a chatbot (And Llama Index can be used with that as well!)

ssunwoong

Thanks for the answer!

According to your answer, if I want to generate some texts with llama-index, then it can be done by using prompt such as "generate a post related to {context}....".

So it would be possible to modify a sentence by referencing a document or embedding stored as an index in llama-index?

LLogan M

Pretty much!

Lets say you had a bunch of poems in an index. Then you could query the index with something like this: index.query("In the style of the existing context, write a poem about snow", similarity_top_k=3)

This is just a simple example though, but I think would work 👍

ssunwoong

Thanks! I think when it generates a poem somehow, I can also see the source of the poem by .get_formatted_sources(), right?

By the way, I was thinking of a system to edit essays based on a specific document DB.

LLogan M

Exactly, that function will show which pieces of text were used to build the generated poem

I think that's a super cool idea btw! 💪

ssunwoong

If I don't plan on using a chat front-end, then llama will do just fine! Thanks.

ssunwoong

Oh, I have a one last question! Does service_context prevent the InvalidRequestError which is occured because of context length? Is this a kind of same role as split text in langchain?

LLogan M

Yea, llama index tries to ensure that everything sent to the LLM is within the max context length. The PromptHelper and ServiceContext objects provide some additional settings for this 👍

https://gpt-index.readthedocs.io/en/latest/guides/primer/usage_pattern.html#customizing-llm-s

ssunwoong

Thanks. It really helps me a lot.

ssunwoong

Hope you have a nice day 🙂

LLogan M

Glad I could help, have a good one too! 👋

Add a reply

Find answers from the community

Hello