I can't tell from documentation if we're meant to use a chat template or if it's automatically implemented?

[sidagarwal2805]

For example at the moment I have a rough chat template:

"[INST] Classify the following text between the delimiters as "normal" or "abnormal" and output your response in JSON format. TEXT: {{{sample_text}}} [/INST] RESPONSE: "

Is this correct usage for Llama/Mistral models or should I not be using them at all?