mistral_inference.py

import requests

def query_mistral(api_url, headers, system_prompt, user_prompt):
    payload = {
        "inputs": f"<s>[INST] {system_prompt}\n\n{user_prompt} [/INST]",
    }

    response = requests.post(api_url, headers=headers, json=payload)
    response.raise_for_status()
    return response.json()["generated_text"]