The best team right there! Absolutely love the crew AI framework + groq lpu speed + llama 3.1 405B open source models! The main limiting factors are stringent rate limits and the super small / limited input and output token limits let's do what chat gpt did with 60,000 tokens in and out per call! That would be truly game changing and launch open source above open AI gpt 4o once and for all! (Greater token & rate limits in and out are what will allow agentic workflows to truly work correctly and as intended for complex processes and architectures. Right now crewAI is great for simple tasks but for longer multi step and complex processes it will need large token limits and ideally limited rate limits!) As always great work team! 😎 🚀 👏🏻👏🏻
Groq is super cool! Good documentation and straightforward UX. I am using the API on daily bases as a VSCode and Inteliij provider for Continue. Insanely fast. Consider to use it as a main provider in the SAAS app, I am currently building. Success to both Groq and Crew AI
Can you convince me to use CrewAI instead of LangGraph with API calls to LLMs and expert systems? I'm concerned about being too limited with CrewAI. Working on an iterative RAG workflow, sort of reasoner for an engineering application. Thoughts?
Groq has incredible inference performance, but I have run into issues with it reliably outputting JSON, which model/LLM with Groq do you recommended for reliable JSON formatted outputs? For multi-step reasoning/chains/agent graphs, I normally have each previous step, output a JSON that feeds into the next step, for the whole chain/graph to not break I need each step doing an LLM call via Groq API, to reliably output JSON.
Actually, figured that using Mixtral "mixtral-8x7b-32768" as the model with Groq produces reliable JSON outputs. I was using LLama earlier. But as a user, I want to be able to use the latest models such as LLama 3.1 as well and get those to output JSON reliably.
Matt on the Groq channel? Hell yeah!
The best team right there! Absolutely love the crew AI framework + groq lpu speed + llama 3.1 405B open source models! The main limiting factors are stringent rate limits and the super small / limited input and output token limits let's do what chat gpt did with 60,000 tokens in and out per call! That would be truly game changing and launch open source above open AI gpt 4o once and for all!
(Greater token & rate limits in and out are what will allow agentic workflows to truly work correctly and as intended for complex processes and architectures. Right now crewAI is great for simple tasks but for longer multi step and complex processes it will need large token limits and ideally limited rate limits!) As always great work team! 😎 🚀 👏🏻👏🏻
Awesome demo! im waiting for that link for jupyter notebook. please share it ! thank you!
yes link would be great!
Sure!! Pleas share the notebook
Groq is super cool! Good documentation and straightforward UX. I am using the API on daily bases as a VSCode and Inteliij provider for Continue. Insanely fast. Consider to use it as a main provider in the SAAS app, I am currently building. Success to both Groq and Crew AI
How are u dealing with free rate limit
Please link the notebooks in the description.
This is awesome! Can we please get a link to the notebook?
Amazing !!!
The problem I have with crew abs groq combo is the timeout limits
Nothing has been linked
Can you convince me to use CrewAI instead of LangGraph with API calls to LLMs and expert systems? I'm concerned about being too limited with CrewAI. Working on an iterative RAG workflow, sort of reasoner for an engineering application. Thoughts?
Groq has incredible inference performance, but I have run into issues with it reliably outputting JSON, which model/LLM with Groq do you recommended for reliable JSON formatted outputs? For multi-step reasoning/chains/agent graphs, I normally have each previous step, output a JSON that feeds into the next step, for the whole chain/graph to not break I need each step doing an LLM call via Groq API, to reliably output JSON.
Actually, figured that using Mixtral "mixtral-8x7b-32768" as the model with Groq produces reliable JSON outputs. I was using LLama earlier. But as a user, I want to be able to use the latest models such as LLama 3.1 as well and get those to output JSON reliably.
Where oh where might we find this impressive notebook file link @GroqInc ? Forever grateful to you and the team!
Don't forget to add the Jupyter Notebook link?
# nice
Best multi agent framework is agency swarm. Why would you make a presentation with hallucinations to promote the product? Haha
so much problems with tool calling
Notebooks are not published not worth watching
Why do you need to use some bloated pile of crap framework though?