Building open source LLM agents with Llama 3 

Great video. Enjoy videos like this that involve building local opensource models, agents, multi agents, tool calls, RAG, etc. The future of AI isn't going to be larger proprietary models, but smaller open ones that are specialized in specific tasks, with a rag database and orchestrated together. Thanks for all the great videos!

how bout multi agent framework and how to achieve 100% reliability and 0% error rate

2nd type if llm, like vlm, as a tool for agent, good idea :) thanks

Tool use without Groq 🤷‍♂️?

Thanks for the video Lance, this actually makes it super clear. I would love to use Groq as llm provider, but does it stores any conversation. Wondering if it provides enterprise level security and compliance?

great !! is it possible to have the same tutorial but with a local model with ollama or llama.cpp?

the link in the description to the source code doesnt exist. Where can we find it?

Great video! But for all of you who will try this, it’s important to emphasize that the official llama3 model will not work for you. What’s being used here is the Groq version which they’ve fine-tuned to support tool calling. I’m just reiterating this because some comments suggests that it’s not clear. Good luck 🍀

Great jobs Lance, and is it depending on llama3? I found sometimes bind_tools output different format or cannot work on some other llms? is it possible to make it universal or else is it possible to provide a list of tested llms? cheers mate!

Could you approximate tool use with the 7b model if you had some sort of format checking of the output?

Thanks for the demo Lance. I don;t want to use any LLM providers, instead I want to use local LLM via Ollama. It seems like Ollama based LLM doesn't support tools. Any idea?

This is really helpful. So the binding action only acts to retrieve the inputs for the tools it understands that need to be used. I believe this can get pretty inaccurate if we ingest 100 tools right? Can you recommend how this should be structured if we would like to have several assistants being responsible for certain actions instead of one assistant deciding which tools to run? Will you ever create a bot to replace yourself so that we can ask these questions to it :D ?

Thanks. How to take two inputs for a function (tool)?

I'm still confuse about Chain and Agent, when I can use Chain, when should I use Agent, is anyone can help me to answer my query, please? Thanks !

Just installed llama3 , it can not remember my name!