import { LattifAIRuntime } from './runtime';
import type { AssetsPathConfig, ContextOptions, LoadModelConfig, LoadedContextInfo, ModelMetadata, LattifAIConfig } from './runtime';
export interface LoadEncryptedBundleOptions {
    /**
     * 加密 ct 来源 — 三选一. 优先级:
     *   1. `ctStream` — 已就绪的 ReadableStream (调用方自己 fetch / file.stream())
     *   2. `ctSource` — Blob | File | URL | string. wrapper 内部 .stream() 或 fetch
     *   3. `ct`       — Uint8Array. 小文件 (<600 MiB) 走累积模式; 大文件自动升级流式
     *
     * 至少必须给一个. 三个都给时按上述优先级取.
     */
    ct?: Uint8Array;
    ctStream?: ReadableStream<Uint8Array>;
    ctSource?: Blob | URL | string;
    /**
     * Optional ct byte size — 仅在流式路径作为 plaintext buffer 上限提示
     * (一次 alloc 省 grow). Blob / response.headers.content-length 命中时
     * 自动填; 直接传 ctStream 时建议手动给.
     */
    ctSize?: number;
    /** End-user license key, e.g. "LAI-XXXX-...". */
    licenseKey: string;
    /** Model id signed onto the license entitlements. */
    modelId: string;
    /**
     * sha256 of `ct`, hex-encoded. /v1/model metadata 返回的 sha256_encrypted.
     *
     * - **ct 路径 (Uint8Array)**: 必填; wrapper 在 decrypt 前做一次性校验,
     *   失败即 fail-fast (CDN tampering / corrupt cache).
     * - **ctStream / ctSource 路径**: 可选 / 忽略. 流式不预读完整 ct 算 sha,
     *   等价的完整性保障由 LaeStream 内部的 chunk 级 HMAC + manifest HMAC
     *   覆盖 (失败抛 ERR_DECRYPT). 提前给也不会用.
     */
    sha256Encrypted?: string;
    /** Forwarded to internal loadModel (n_ctx, n_batch, kv types, …). */
    loadConfig?: LoadModelConfig;
    /** Forwarded to internal loadMmproj when an encoder segment is present. */
    mmprojOpts?: {
        useGpu?: boolean;
        nThreads?: number;
    };
}
export interface MmprojCaps {
    supportsAudio: boolean;
    supportsVision: boolean;
    usesMrope: boolean;
    audioSampleRate: number;
}
export interface LoadEncryptedBundleResult {
    /**
     * Encoder capabilities (only meaningful when the bundle ships an
     * encoder segment; otherwise all-false / 0).
     */
    caps: MmprojCaps;
}
export declare class LattifAI {
    #private;
    constructor(paths: AssetsPathConfig, config?: LattifAIConfig);
    loadEncryptedBundle(opts: LoadEncryptedBundleOptions): Promise<LoadEncryptedBundleResult>;
    isModelLoaded(): boolean;
    getModelMetadata(): ModelMetadata;
    getLoadedContextInfo(): LoadedContextInfo;
    getBOS(): number;
    getEOS(): number;
    getEOT(): number;
    isTokenEOG(t: number): boolean;
    isMultithread(): boolean;
    usingWebGPU(): boolean;
    getNumThreads(): number;
    getChatTemplate(): string | null;
    isEncoderDecoderArchitecture(): boolean;
    mustAddBosToken(): boolean;
    mustAddEosToken(): boolean;
    tokenize(text: string, special?: boolean): Promise<number[]>;
    detokenize(tokens: number[], returnString?: boolean): Promise<Uint8Array | string>;
    getVocab(): Promise<Uint8Array[]>;
    lookupToken(piece: string): Promise<number>;
    samplingInit(...args: Parameters<LattifAIRuntime['samplingInit']>): Promise<void>;
    samplingSample(): Promise<{
        piece: Uint8Array;
        token: number;
    }>;
    samplingAccept(tokens: number[]): Promise<void>;
    getLogits(topK?: number): Promise<{
        token: number;
        p: number;
    }[]>;
    decode(...args: Parameters<LattifAIRuntime['decode']>): Promise<{
        nPast: number;
    }>;
    encode(...args: Parameters<LattifAIRuntime['encode']>): Promise<{
        nPast: number;
    }>;
    embeddings(tokens: number[]): Promise<number[]>;
    loopDecode(opts: Parameters<LattifAIRuntime['loopDecode']>[0]): Promise<{
        tokens: number[];
        pieces: Uint8Array;
        pieceOffsets: number[];
        stoppedByEog: boolean;
        nPast: number;
    }>;
    mtmdEvalAudio(...args: Parameters<LattifAIRuntime['mtmdEvalAudio']>): Promise<{
        nPast: number;
        nChunks: number;
    }>;
    kvRemove(nKeep: number, nDiscard: number): Promise<void>;
    kvClear(): Promise<void>;
    createCompletion(...args: any[]): any;
    createChatCompletion(...args: any[]): any;
    createEmbedding(...args: Parameters<LattifAIRuntime['createEmbedding']>): Promise<number[]>;
    formatChat(...args: Parameters<LattifAIRuntime['formatChat']>): Promise<string>;
    setOptions(opt: ContextOptions): Promise<void>;
    getPerfContext(): Promise<import("./runtime").PerfContextData>;
    resetPerfContext(): Promise<{
        success: boolean;
    }>;
    exit(): Promise<void>;
}