mlx-bun library API
    Preparing search index...

    Interface ExecutionGroup

    A backend execution group exposes bounded work units and readiness. Queue payloads, tensors, cache merging and sampling remain inside its adapter.

    interface ExecutionGroup {
        active: number;
        admissionHeld: boolean;
        canPrepareMore?: boolean;
        closed: boolean;
        maxActive: number;
        maxIterationTokens?: number;
        maxPreparationTokens?: number;
        mixedPreparation?: {
            minimumPreparationTokens: number;
            runningTokens: number;
        };
        nextPreparationTokens?: number;
        pendingTasks?: number;
        preparationPublishedOutput?: boolean;
        preparing: boolean;
        preparingRows?: number;
        preparingTokens?: number;
        queued: number;
        acquireExecution?(): Promise<() => void>;
        admitNext(): boolean;
        advance(): Promise<void>;
        advanceMixed?(tokenBudget: number): Promise<void>;
        advancePreparation(): Promise<void>;
        advanceTask?(): Promise<void>;
        canBurst(): boolean;
        failActive(error: unknown): void;
        failAll(error: unknown): void;
        pruneCancelled(): void;
        reserveResidency(): () => void;
        waitForWork(): Promise<void>;
    }
    Index
    active admissionHeld canPrepareMore? closed maxActive maxIterationTokens? maxPreparationTokens? mixedPreparation? nextPreparationTokens? pendingTasks? preparationPublishedOutput? preparing preparingRows? preparingTokens? queued
    acquireExecution? admitNext advance advanceMixed? advancePreparation advanceTask? canBurst failActive failAll pruneCancelled reserveResidency waitForWork
    active: number
    admissionHeld: boolean
    canPrepareMore?: boolean

    The preparation owner can accept another row before its next work unit.

    closed: boolean
    maxActive: number
    maxIterationTokens?: number
    maxPreparationTokens?: number
    mixedPreparation?: { minimumPreparationTokens: number; runningTokens: number }

    Method-owned token demand for a compatible mixed iteration. Absent when the backend has no packed implementation for the current work.

    nextPreparationTokens?: number
    pendingTasks?: number

    Native preparation may borrow the execution lease between work units. Payloads and task failures belong to the backend's queue.

    preparationPublishedOutput?: boolean

    The latest preparation published output that can be flushed by yielding.

    preparing: boolean
    preparingRows?: number
    preparingTokens?: number

    Work estimates affect grouping only. A request larger than this budget still runs alone; it is never rejected or sent to another executor.

    queued: number

    Settings
    Member Visibility

    On This Page
    Properties
    Methods