Question about Llama 3.2 3b instruct, multi-shot prompts and how to represent examples

[andrewfr]

Hi Folks:

I am new to llama3.2 and prompting. As the title says, I am interested what the proper format is for examples using
multi-shot prompting (or fine-tuning) involving tool calling. Or if it is possible?

I have been using llama3.2 with ollama.chat(). I am tool calling. I am giving what the ollama
3.2 3b instruct template expects (i.e., a list of tools in JSON format). To check accuracy/consistency, I run a test 30 times and examine the results. The results are all over the place. I am hoping I could provide a few examples. I would expect the examples to be a part of role:assistant and the property being tools_calls. Or would I be using role:ipython? Any advice would be appreciated!

Cheers,
Andrew