Mind Blowing Function Calling by New Hermes 2 on Llama 3 Locally

This video is a hands-on step-by-step tutorial to use function calling and tools locally using phidata and ollama with streamlit.
🔥 Buy Me a Coffee to support the channel: ko-fi.com/fahd...
▶ Become a Patron 🔥 - / fahdmirza
#phidata #streamlit #ollama #llama3 #openhermes2
PLEASE FOLLOW ME:
▶ LinkedIn: / fahdmirza
▶ TH-cam: / @fahdmirza
▶ Blog: www.fahdmirza.com
RELATED VIDEOS:
▶ Phidata www.phidata.com/
All rights reserved © 2021 Fahd Mirza

[@szpiegzkrainydeszczowcow8476]

Any chance on video on building long term memory with local Llama3? Greetings!

[@truliapro7112]

I want to see that too.

[@fahdmirza]

Sure, please give bit more info on this and I will try my best, thanks.

[@szpiegzkrainydeszczowcow8476]

@@fahdmirza how about using agentic RAG that is embedding history of conversation in JSON and then take it as context everytime you speak with it. One agent decide to summiraze conversation and write in history of chat while other is looking for important context when in conversation with you. this would give long term memory to local AI and made it more personal. Maybe there is even better way to do it?

[@pin65371]

@@szpiegzkrainydeszczowcow8476 I've thought that would be a good process after a reflection step once a task has been completed. You could even just run everything through a private llm at night to have an agent that looks at the entire process to see what it could learn. Then as you said add it to json. You could even put a bunch of metadata in to make it easy to search. Once you get enough data you could do a LoRa fine tune for certain tasks to create agents that are experts in specific tasks. It would also be possible to do A/B tests where you run the agent through a task without the dataset and then again referencing the dataset to see how much an improvement there is. I dont know how efficient it would be for the first little while since you'd be adding costs for tasks by adding extra steps but in the long term I think it would improve quality and possibly costs.
One other thing is if you are doing a variety of tasks it would be possible to start putting everything into knowledge graphs. With a knowledge graph you might start finding ways to connect things in less obvious ways.

[@excido7107]

@@szpiegzkrainydeszczowcow8476 or implementing it nicely with MemGPT? but yes any sort of LTM system would be amazing :)

[@dr.mikeybee]

Nice vide!

[@fahdmirza]

Thank you, much appreciated.

[@themax2go]

i wish you'd have talked a bit at start what it can do and what it can't do, in terms of function calling, specific use cases outside of what you've shown, and how to implement those. maybe an idea for a future video

[@fahdmirza]

Sure, I already have talked about Functions calling a lot in other videos. Please search the channel, thanks.

[@zzzzzzz8473]

thanks for the walkthrough , very exciting ! Hermes is the greek god of heralds , a messenger , pronounced HuuuRmiiiiiis

[@fahdmirza]

Thanks for the info!

[@themax2go]

ai youtuber and actor, model, plastic surgeon, sculptor?!! 🤯💯

[@fahdmirza]

lol cheers

[@MrSuntask]

Love this!

[@fahdmirza]

Thanks.