mlx-bun library API
    Preparing search index...
    interface WhisperDecodingOptions {
        audioCtx?: number | null;
        beamSize?: number | null;
        bestOf?: number | null;
        fast?: boolean;
        language?: string | null;
        lengthPenalty?: number | null;
        maxInitialTimestamp?: number | null;
        observer?: WhisperDecodeObserver;
        patience?: number | null;
        prefix?: string | number[] | null;
        prompt?: string | number[] | null;
        sampleLen?: number | null;
        suppressBlank?: boolean;
        suppressTokens?: number[] | "-1" | null;
        task?: WhisperTask;
        temperature?: number;
        withoutTimestamps?: boolean;
    }

    Hierarchy (View Summary)

    Index
    audioCtx?: number | null

    Lab (whisper.cpp -ac): encode only the first audioCtx of the 1500 encoder positions (2·audioCtx mel frames). Cheaper for short clips but NOT the trained context — measured hallucinated repeats below ~1024 on the fixture clips. null = full context. Fast path only.

    beamSize?: number | null
    bestOf?: number | null
    fast?: boolean

    Execution path: true (default) runs the optimized graphs (the graph's fast path) for temperature-0 decoding; false forces the faithful oracle graph. Sampling (temperature > 0) always uses the faithful path.

    language?: string | null

    null → detect.

    lengthPenalty?: number | null
    maxInitialTimestamp?: number | null

    Parity instrumentation: the decode window's encoder output and each step's PRE-filter last-position logits [n_group, V] f32 (borrowed — valid only during the callback).

    patience?: number | null
    prefix?: string | number[] | null
    prompt?: string | number[] | null
    sampleLen?: number | null
    suppressBlank?: boolean
    suppressTokens?: number[] | "-1" | null

    Token ids to suppress; "-1" = the tokenizer's non-speech set (default).

    temperature?: number
    withoutTimestamps?: boolean