mlx-bun library API
    Preparing search index...

    Interface BenchmarkComparison

    The paired comparison of two runs.

    interface BenchmarkComparison {
        base: string;
        candidate: string;
        comparable: boolean;
        markdown: string;
        tasks: readonly {
            base: number | null;
            candidate: number | null;
            delta: number | null;
            task: string;
        }[];
    }
    Index
    base candidate comparable markdown tasks
    base: string
    candidate: string
    comparable: boolean

    The runner's verdict: both runs complete and comparable. Differences are never accepted by a tolerance.

    markdown: string

    eval-serve.ts compare's report: every score difference, sample flip, skip and provenance mismatch.

    tasks: readonly {
        base: number | null;
        candidate: number | null;
        delta: number | null;
        task: string;
    }[]

    Accuracy per task in either run.

    Settings
    Member Visibility

    On This Page
    Properties