Chat template was missing tool_use from Hermes. If this should be the same then this will do. Otherwise it will need to be reworked for llama 3.1 as the base training on tool calling uses other tokens like <|python_tag|> or <|eom_id|>.