Find answers from the community

Updated 8 months ago

Confused with RAG implementation and its

Confused with RAG implementation and its variants

I’m dealing with a set of texts(first it might be english but aim to expand it later) and need to implement an efficient retrieval system. Do you have any tips on using RAG for this purpose? Also, if there are any updates or best practices specific to multilingual scenarios, I'd love to hear about those.

Any helpful links or resources are welcome!

6 comments

TTeemu

What kind of texts are you working with?

For multilingual: there are some strategies like fine-tuning embeddings:
https://docs.llamaindex.ai/en/stable/examples/finetuning/embeddings/finetune_embedding/?h=embeddings+fine

I've been using OpenAI's models for domain-specific context in another language and the performance has still been good so it might not be necessary.

BBexboy

oh nice i mean same domain specific like mostly historic documents and what about just english what is the best for that?

TTeemu

You can probably start with regular semantic search

TTeemu

But it does really depend on the exact structure of the documents

BBexboy

sure I will definately do btw should i use llama index or langchain? and advice is nice

TTeemu

I've been using Llamaindex basically since it released, including 10+ development projects for clients; and it has always been great. So I'd definitely pick Llamaindex 🦙

Add a reply