Please provide some guidance or documentation on tool usage.
Firstly, thanks so much for your great work. This is a truly brilliant model for its size.
I think there is a blind spot in the documentation. It is clear that Llama 3 has been fine-tuned to use tools. It is able to generate images and search the web as part of the meta.ai implementation. Looking at the tokenizer, there is a /tool token.
It would be greatly appreciated if you could provide some example prompts for defining a custom tool for Llama 3 to use, as this would allow the community to augment the model with our own use-case specific tools (EG: Agentic Retrieval Augmented Generation), which is extremely valuable.
I expect that the model was fine-tuned to use tools defined in a specific format, so knowing this format is very important to enabling tools.