import type { LlamaCppMemorySafetyConfig, LlamaCppModelMemoryInfo } from "./llama-cpp-provider-config.js";
export interface EstimateMemoryUsageOptions {
    model: LlamaCppModelMemoryInfo;
    contextSize: number;
    /**
     * GGUF model file size, if known.
     */
    modelFileSizeBytes?: number;
    /**
     * Multimodal projector file size, if known.
     */
    mmprojFileSizeBytes?: number;
    /**
     * Override bytes per K/V cache element. Defaults to model metadata or f16.
     */
    kvCacheBytesPerValue?: number;
    /**
     * Extra overhead for llama.cpp compute buffers and Metal allocations.
     * Defaults to max(1 GiB, 15% of KV cache).
     */
    computeOverheadBytes?: number;
}
export interface MemoryUsageEstimate {
    contextSize: number;
    kvCacheBytes: number;
    modelBytes: number;
    mmprojBytes: number;
    computeOverheadBytes: number;
    totalBytes: number;
    bytesPerToken: number;
    maxContextSize?: number;
}
export interface MemorySafetyCheckOptions {
    model?: LlamaCppModelMemoryInfo;
    contextSize?: number;
    modelFileSizeBytes?: number;
    mmprojFileSizeBytes?: number;
    memorySafety?: LlamaCppMemorySafetyConfig;
}
export interface MemorySafetyCheckResult {
    contextSize: number;
    estimate?: MemoryUsageEstimate;
}
export declare function estimateMemoryUsage(options: EstimateMemoryUsageOptions): MemoryUsageEstimate;
export declare function checkMemorySafety(options: MemorySafetyCheckOptions): MemorySafetyCheckResult;
//# sourceMappingURL=memory-estimation.d.ts.map