Llama 3 tool calling agents with Firefunction-v2

preview_player
Показать описание
Firefunction-v2 is a new open weights model from Fireworks fine-tuned specifically for tool calling. Built on Llama 3 70b, Firefunction-v2 retains strong instruction following capabilities with tool calling on par with gpt4-o!

We show how to use this model and, importantly, benchmark its tool calling ability for yourself. We test Firefunction-v2 on a small SQL agent challenge, which shows performance on par with gpt4-o.

Using Fireworks:

Agent evaluation cookbook:

Agent evaluation video:
Рекомендации по теме
Комментарии
Автор

Literally, I was dealing with llama3 and integrating tools within it...got many errors...
And now I just got this video

pragyantiwari
Автор

Few days back tried a lot.. finally went with gemini model. Worked fantastically.

sabre_code
Автор

Hi Lance - great video again!Question for you. Recently I’ve been omitting LLM function calling just as a precautionary measure. I’m basically separating out the LLM from my functions (like api calls) and just asking the LLM to return a jsonOutput compliant with the parms in the api function. Am I doing myself a disservice by keeping these separate?

jellz
Автор

keep doing you are doing great 🎉🎉🎉🎉❤❤❤

darkmatter
Автор

For everything if you say "you can look into the notebook" why put up a video? The video is running as if being chased by someone.

sravan
Автор

but it doesn't seem like firework is free.

wryltxw