ai-sdk-ollama/dist/models/embedding-reranking-model.d.ts

Vercel AI SDK Provider for Ollama using official ollama-js library

import { RerankingModelV4, RerankingModelV4CallOptions } from '@ai-sdk/provider';
import { Ollama } from 'ollama';
/**
 * Configuration for the Ollama embedding-based reranking model
 */
export interface OllamaEmbeddingRerankingConfig {
    client: Ollama;
    provider: string;
}
/**
 * Settings for configuring Ollama embedding-based reranking
 */
export interface OllamaEmbeddingRerankingSettings {
    /**
     * Embedding model to use for computing document similarity.
     * If not specified, uses the modelId passed to the constructor.
     * Recommended models: 'bge-m3', 'nomic-embed-text', 'mxbai-embed-large'
     */
    embeddingModel?: string;
    /**
     * Maximum number of texts to embed per request. Smaller batches reduce
     * memory/latency spikes for large document sets while still avoiding one
     * request per document. Defaults to 16.
     */
    maxBatchSize?: number;
}
/**
 * Embedding-Based Reranking Model (Workaround)
 *
 * Since Ollama doesn't have native reranking support yet (PR #11389 not merged),
 * this implementation uses embedding similarity as a workaround:
 *
 * 1. Embed the query using an embedding model
 * 2. Embed all documents using the same model
 * 3. Calculate cosine similarity between query and each document
 * 4. Sort documents by similarity score (descending)
 *
 * This approach works with any Ollama embedding model and provides
 * reasonable reranking results for most use cases.
 *
 * @example
 * ```ts
 * import { ollama } from 'ai-sdk-ollama';
 * import { rerank } from 'ai';
 *
 * const result = await rerank({
 *   model: ollama.embeddingReranking('bge-m3'),
 *   query: 'What is machine learning?',
 *   documents: [
 *     'Machine learning is a subset of AI...',
 *     'The weather today is sunny...',
 *     'Deep learning uses neural networks...',
 *   ],
 *   topN: 2,
 * });
 *
 * console.log(result.rerankedDocuments);
 * // Documents sorted by relevance to the query
 * ```
 */
export declare class OllamaEmbeddingRerankingModel implements RerankingModelV4 {
    readonly specificationVersion: 'v4';
    readonly modelId: string;
    private readonly config;
    private readonly settings;
    constructor(modelId: string, settings: OllamaEmbeddingRerankingSettings, config: OllamaEmbeddingRerankingConfig);
    get provider(): string;
    /**
     * Get the effective embedding model to use
     */
    private get embeddingModelId();
    /**
     * Normalized batch size for embedding requests. Ensures we never request
     * non-positive batch sizes.
     */
    private get embeddingBatchSize();
    /**
     * Embed a batch of texts while keeping their order aligned with the
     * embeddings that are returned.
     */
    private embedBatch;
    doRerank({ documents, query, topN, }: RerankingModelV4CallOptions): Promise<Awaited<ReturnType<RerankingModelV4['doRerank']>>>;
}
//# sourceMappingURL=embedding-reranking-model.d.ts.map