/**
 * Paged KV Cache implementation
 */
import type { Tensor } from './tensor.ts';
/**
 * Paged KV Cache for efficient memory management
 *
 * Layout: [num_pages, 2, page_size, num_kv_heads, head_dim]
 * where dim 1 has [0] = keys, [1] = values
 */
export interface PagedKvCache {
    /**
     * The backing tensor for the paged KV cache
     * Shape: [num_pages, 2, page_size, num_kv_heads, head_dim]
     */
    data: Tensor;
    /**
     * Number of tokens per page
     */
    page_size: number;
    /**
     * Number of KV heads (for GQA/MQA)
     */
    num_kv_heads: number;
    /**
     * Dimension of each head
     */
    head_dim: number;
}
/**
 * Create a paged KV cache
 */
export declare function createPagedKvCache(data: Tensor, page_size: number, num_kv_heads: number, head_dim: number): PagedKvCache;