Fix typo in the Llama3.1 prompt template

[wizzard0]

The llama 3.1 tool calling template says "orginal user question" instead of the "original user question".

Unless it's the actual prompt pattern the model is trained with, this should affect the tool use performance; and I see this typo spreading to other tutorials and tools using Llama 3.1

[pcuenca]

Hi @wizzard0! Actually, the typo is also present in the official documentation (see Step 3 in the JSON-based tool calling section). I doubt that the system prompt has to be supplied verbatim like that and I'd prefer to have the typo fixed, but I haven't really tested it comprehensively. Have you observed differences one way or another?