Feature request: --output json flag for machine-readable results

[b-gullekson]

Feature request: --output json flag for machine-readable results

Currently ollama-benchmark outputs results in a human-readable text format (or table with -t). It would be really useful to have a --output json option that writes structured results to stdout or a file.

Use case: I'm benchmarking models across several machines and want to collect results programmatically — aggregate them in a spreadsheet, compare across hardware, or feed into a monitoring dashboard. Right now I'd need to parse the text output which is fragile.

Possible approach: The OllamaResponse objects already use Pydantic models with all the metrics, so serializing to JSON should be straightforward — something like:

if args.output == "json":
    results = []
    for model_name, responses in benchmarks.items():
        results.append({
            "model": model_name,
            "runs": [r.model_dump() for r in responses]
        })
    print(json.dumps(results, indent=2, default=str))

Happy to take a crack at this if it sounds useful.

[LarHope]

This is a great idea — JSON output would make it much easier to compare results across machines. The Pydantic models do make this straightforward. I'll look into adding this, probably as --output json alongside the existing text and table modes. Thanks for the suggestion and the code sketch!