How To Use Different Vectorstores | Llama2 With LangChain | Chat With Documents | Chainlit

There are many vectorstores and one can use any of them based on the use cases they have. In this video, I will demonstrate how you can use ChromaDB, Faiss and Pinecone to store embeddings and retrieve that for creating a chatbot to chat with documents. The main idea here is to use different vectorstores rather than the final output which might depend upon the machine you use.

What I noticed from the experiment is that Llama2 model is not that good compared to OpenAI model for QA Retrieval part. It might depend upon your machine / hardware. All the mentioned links are either in this description or in the readme file of the github repository.

Happy Learning 😎

👉🏼 Links:

⏰ Timestamps:
00:00 Introduction
01:36 Setup
06:34 ChromaDB with Llama2 (ingest + app)
18:32 FAISS with Llama2 (ingest + app)
23:40 Pinecone Llama2 (ingest + app)
33:56 ChromDB with OpenAI model (ingest + app)
39:14 Conclusion

------------------------------------------------------------------------------------------

------------------------------------------------------------------------------------------
🔗 🎥 Other videos you might find helpful:

------------------------------------------------------------------------------------------
🤝 Connect with me:

#llama2 #llama #chainlit #langchain #llm #chatwithpdf #chromadb #datasciencebasics

[mdaslamknl]

Excellent


Keep it up

Thank You

[VenkatesanVenkat-fdhg]

Thanks for your valuable videos. Can you discuss llama 2 sagemaker deployment or llama 2 + prompt engineering..

[tamilil-]

Hi .. excellent work .. thanks a lot Sir.. But I tried to run the project, but it threw a `TypeError: expected string or buffer` error. I think that's because of the package versions but I am clueless. Can you please give a solution for this?

[myiybmk]

I'm a newcomer to the world of Language Model (LLM), and I've been watching your videos to learn. I've got a question: Is the Hugging Face API free, or is there a cost involved, similar to OpenAI?

[henkhbit]

thanks for the comparison. Strange I see other video where the llma2 7b 4 bits model gave correct result with a pdf document... Has it to do with too little ram available? btw: I saw in this video that sometimes the answer of the chatbot appears twice...

[DanielWeikert]

Why did you chose 384 dimesion size for the vectorstore for lama2 ? Are different sizes possible?
Why did you choose alll_minLLM-L6-v2 vor Huggingface Ebeddings? Other models possible?
Thx

[zgvuwtv]

which one performs best in terms of vector db ?
1) pinecone
2)chromadb
3)faiss
???? give me rating any idea would highly appreciated if help me rate them in order

[vsudbdk]

How to overcome the token excessive error (like 1k tokens excessive than 512 tokens like that), I have been trying to deal with it. If I perform truncation of relevant doc result I may loose the necessary answer which might lie after the truncated section of information, I also tried to reduce the k clusters to 2 but still not much change. Can you suggest a way to overcome it.

[anuragbhandari]

hey, i'm trying to replicate the pinecone implementation that you've done, but instead of the normal llama-2 model, im using a fine-tuned llama-2 model but when i run the chainlit app, its returning this error that "Exception: Failed to load model: Failed to create LLM "model name" from 'model path'", the error is traced back to the line where we are creating the llm by using the CTransformers

[deejay-mv]

Can you do another video on using pgvector?