Fine-tuning a model with JSON/dictionary as a prompt

I am trying to create a simple file with a number of prompts and completions for fine-tuning. Essentially, I want to convert a JSON object with specific data points to a generated narrative, along the lines of:

prompt = {
        "metric": "AR Balance",
        "type": "attribution",
        "period": "May '23",
        "period_mom": "Apr '23",
        "period_qoq": "Feb '23",
        "period_yoy": "May '22",
        "portfolio": "Big 6",
        "portfolio_region": "US/CA",
        "value": 196807288,
        "value_mom": 90753765,
        "value_qoq": 80792641,
        "delta_mom": 106053523,
        "delta_qoq": 116014647,
        "delta_pct_mom": 1.16,
        "delta_pct_qoq": 1.43,
        "score_mom": 91.7,
        "score_qoq": 91.7,
}

completion = " The portfolio Big6 in US/CA experienced a notable increase to $196.8M, representing a ▲ 116% ($106.1M) MoM change"

However, I am struggling how to create a simple fine-tuning file, using JSON as an input in the prompt as it thinks it’s an extension of the prompt/completion JSON structure. Any suggestions?

{
    "prompt": "add JSON here ->", 
    "completion": "generated narrative\n"
}
{
    "prompt": "add JSON here ->", 
    "completion": "generated narrative\n"
}