/**
 * Runner Types
 * Types for communicating with the llama.cpp runner process
 */
import type { ModelOptions } from './api.js';
export declare enum RunnerStatus {
    STARTING = "starting",
    LOADING = "loading",
    READY = "ready",
    BUSY = "busy",
    ERROR = "error",
    STOPPED = "stopped"
}
export interface RunnerProcess {
    pid: number;
    port: number;
    status: RunnerStatus;
    modelPath: string;
    startedAt: Date;
    lastUsedAt: Date;
    vramSize: number;
    totalSize: number;
}
export interface CompletionRequest {
    prompt: string;
    images?: string[];
    grammar?: string;
    cachePrompt?: boolean;
    temperature?: number;
    topK?: number;
    topP?: number;
    minP?: number;
    typicalP?: number;
    repeatPenalty?: number;
    repeatLastN?: number;
    presencePenalty?: number;
    frequencyPenalty?: number;
    seed?: number;
    stop?: string[];
    nPredict?: number;
    nKeep?: number;
    logprobs?: boolean;
    topLogprobs?: number;
    stream?: boolean;
}
export interface CompletionResponse {
    content: string;
    stop: boolean;
    stopReason?: 'stop' | 'length' | 'eos';
    generationSettings?: Record<string, unknown>;
    model?: string;
    promptEvalCount?: number;
    promptEvalDuration?: number;
    evalCount?: number;
    evalDuration?: number;
    completionProbabilities?: TokenProbability[];
}
export interface TokenProbability {
    content: string;
    probs: Array<{
        tok_str: string;
        prob: number;
    }>;
}
export interface EmbeddingRunnerRequest {
    content: string;
}
export interface EmbeddingRunnerResponse {
    embedding: number[];
    promptEvalCount: number;
}
export interface TokenizeRequest {
    content: string;
}
export interface TokenizeResponse {
    tokens: number[];
}
export interface DetokenizeRequest {
    tokens: number[];
}
export interface DetokenizeResponse {
    content: string;
}
export interface HealthResponse {
    status: 'ok' | 'loading' | 'error';
    progress?: number;
    slotsIdle?: number;
    slotsProcessing?: number;
}
export interface LoadRequest {
    modelPath: string;
    projectorPath?: string;
    adapterPaths?: string[];
    options?: ModelOptions;
    systemInfo?: SystemInfo;
    gpus?: GPUInfo[];
}
export interface SystemInfo {
    totalMemory: number;
    freeMemory: number;
    cpuCount: number;
    platform: string;
    arch: string;
}
export interface GPUInfo {
    id: string;
    name: string;
    vendor: string;
    vram: number;
    freeVram: number;
    computeCapability?: string;
}
export type DeviceType = 'cpu' | 'cuda' | 'metal' | 'rocm' | 'vulkan';
export interface LayerAllocation {
    deviceId: string;
    deviceType: DeviceType;
    layers: number;
    vramUsed: number;
}
export interface RunnerConfig {
    runnerPath?: string;
    modelsPath: string;
    host: string;
    port: number;
    maxLoadedModels: number;
    maxVram?: number;
    gpuOverhead: number;
    loadTimeout: number;
    requestTimeout: number;
    keepAliveTimeout: number;
    flashAttention: boolean;
    kvCacheType: 'f16' | 'q8_0' | 'q4_0';
    cudaVisibleDevices?: string;
    rocmVisibleDevices?: string;
}
export declare const DEFAULT_RUNNER_CONFIG: RunnerConfig;
//# sourceMappingURL=runner.d.ts.map