openvino-genai-node/types/addon.d.ts

OpenVINO™ GenAI pipelines for using from Node.js environment

import { Tensor } from "openvino-node";
import type { ChatHistory as IChatHistory } from "./chatHistory.js";
import type { Tokenizer as ITokenizer } from "./tokenizer.js";
import { IReasoningParser, IDeepSeekR1ReasoningParser, IPhi4ReasoningParser, ILlama3PythonicToolParser, ILlama3JsonToolParser } from "./parsers.js";
import { GenerationConfig, GenerationFinishReason, StreamingStatus, VLMPipelineProperties, LLMPipelineProperties, WhisperGenerationConfig, WhisperPipelineProperties, SpeechGenerationConfig, ImageGenerationConfig, ImageGenerationCallback, Text2ImagePipelineProperties, Image2ImagePipelineProperties, InpaintingPipelineProperties, Text2SpeechPipelineProperties } from "./utils.js";
import { VLMPerfMetrics, PerfMetrics, WhisperPerfMetrics, ImageGenerationPerfMetrics, Text2SpeechPerfMetrics } from "./perfMetrics.js";
import type { WhisperDecodedResultChunk, WhisperWordTiming } from "./decodedResults.js";
export type EmbeddingResult = Float32Array | Int8Array | Uint8Array;
export type EmbeddingResults = Float32Array[] | Int8Array[] | Uint8Array[];
export type TextRerankResult = [index: number, score: number];
export type TextRerankResults = TextRerankResult[];
/**
 * Pooling strategy
 */
export declare enum PoolingType {
    /** First token embeddings */
    CLS = 0,
    /** The average of all token embeddings */
    MEAN = 1
}
export type TextEmbeddingConfig = {
    /** Maximum length of tokens passed to the embedding model */
    max_length?: number;
    /** If 'true', model input tensors are padded to the maximum length */
    pad_to_max_length?: boolean;
    /** Side to use for padding "left" or "right" */
    padding_side?: "left" | "right";
    /**
     * Batch size of embedding model.
     * Useful for database population. If set, the pipeline will fix model shape for inference optimization.
     * Number of documents passed to pipeline should be equal to batch_size.
     * For query embeddings, batch_size should be set to 1 or not set.
     */
    batch_size?: number;
    /** Pooling strategy applied to model output tensor */
    pooling_type?: PoolingType;
    /** If 'true', L2 normalization is applied to embeddings */
    normalize?: boolean;
    /** Instruction to use for embedding a query */
    query_instruction?: string;
    /** Instruction to use for embedding a document */
    embed_instruction?: string;
};
export interface TextEmbeddingPipelineWrapper {
    new (): TextEmbeddingPipelineWrapper;
    init(modelPath: string, device: string, config: TextEmbeddingConfig, ovProperties: object, callback: (err: Error | null) => void): void;
    embedQuery(text: string, callback: (err: Error | null, value: EmbeddingResult) => void): void;
    embedDocuments(documents: string[], callback: (err: Error | null, value: EmbeddingResults) => void): void;
    embedQuerySync(text: string): EmbeddingResult;
    embedDocumentsSync(documents: string[]): EmbeddingResults;
}
/**
 * Configuration parameters for TextRerankPipeline.
 */
export type TextRerankPipelineConfig = {
    /**
     * Number of documents to return sorted by score.
     * @defaultValue 3
     */
    top_n?: number;
    /** Maximum length of tokens passed to the embedding model. */
    max_length?: number;
    /** If 'true', model input tensors are padded to the maximum length. */
    pad_to_max_length?: boolean;
    /** Side to use for padding "left" or "right". */
    padding_side?: "left" | "right";
};
export interface TextRerankPipeline {
    new (): TextRerankPipeline;
    init(modelPath: string, device: string, config: TextRerankPipelineConfig, ovProperties: object, callback: (err: Error | null) => void): void;
    rerank(query: string, documents: string[], callback: (err: Error | null, value: TextRerankResults) => void): void;
}
export interface LLMPipeline {
    new (): LLMPipeline;
    init(modelPath: string, device: string, ovProperties: LLMPipelineProperties, callback: (err: Error | null) => void): void;
    generate(inputs: string | string[] | IChatHistory, generationConfig: GenerationConfig, streamer: ((chunk: string) => StreamingStatus) | undefined, callback: (err: Error | null, result: {
        texts: string[];
        scores: number[];
        perfMetrics: PerfMetrics;
        parsed: Record<string, unknown>[];
        finishReasons: GenerationFinishReason[];
    }) => void): void;
    startChat(systemMessage: string, callback: (err: Error | null) => void): void;
    finishChat(callback: (err: Error | null) => void): void;
    getTokenizer(): ITokenizer;
    getGenerationConfig(): GenerationConfig;
    setGenerationConfig(config: GenerationConfig): void;
}
export interface WhisperPipeline {
    new (): WhisperPipeline;
    init(modelPath: string, device: string, properties: WhisperPipelineProperties, callback: (err: Error | null) => void): void;
    generate(rawSpeech: Float32Array | number[], generationConfig: WhisperGenerationConfig, streamer: ((chunk: string) => StreamingStatus) | undefined, callback: (err: Error | null, result: {
        texts: string[];
        scores: number[];
        perfMetrics: WhisperPerfMetrics;
        chunks?: WhisperDecodedResultChunk[];
        words?: WhisperWordTiming[];
    }) => void): void;
    getTokenizer(): ITokenizer;
    getGenerationConfig(): Partial<WhisperGenerationConfig>;
    setGenerationConfig(config: WhisperGenerationConfig): void;
}
export interface VLMPipeline {
    new (): VLMPipeline;
    init(modelPath: string, device: string, ovProperties: VLMPipelineProperties, callback: (err: Error | null) => void): void;
    generate(inputs: string | IChatHistory, images: Tensor[] | undefined, videos: Tensor[] | undefined, streamer: ((chunk: string) => StreamingStatus) | undefined, generationConfig: GenerationConfig | undefined, callback: (err: Error | null, result: {
        texts: string[];
        scores: number[];
        perfMetrics: VLMPerfMetrics;
        parsed: Record<string, unknown>[];
        finishReasons: GenerationFinishReason[];
    }) => void): void;
    startChat(systemMessage: string, callback: (err: Error | null) => void): void;
    finishChat(callback: (err: Error | null) => void): void;
    getTokenizer(): ITokenizer;
    setChatTemplate(template: string): void;
    setGenerationConfig(config: GenerationConfig): void;
    getGenerationConfig(): GenerationConfig;
}
export interface Text2ImagePipeline {
    new (): Text2ImagePipeline;
    init(modelPath: string, device: string, properties: Text2ImagePipelineProperties, callback: (err: Error | null) => void): void;
    generate(prompt: string, properties: ImageGenerationConfig, streamer: ImageGenerationCallback | undefined, callback: (err: Error | null, result: Tensor) => void): void;
    decode(latent: Tensor, callback: (err: Error | null, result: Tensor) => void): void;
    getPerformanceMetrics(): ImageGenerationPerfMetrics;
    getGenerationConfig(): ImageGenerationConfig;
    setGenerationConfig(config: ImageGenerationConfig): void;
}
export interface Image2ImagePipeline {
    new (): Image2ImagePipeline;
    init(modelPath: string, device: string, properties: Image2ImagePipelineProperties, callback: (err: Error | null) => void): void;
    generate(prompt: string, image: Tensor, properties: ImageGenerationConfig, streamer: ImageGenerationCallback | undefined, callback: (err: Error | null, result: Tensor) => void): void;
    decode(latent: Tensor, callback: (err: Error | null, result: Tensor) => void): void;
    getPerformanceMetrics(): ImageGenerationPerfMetrics;
    getGenerationConfig(): ImageGenerationConfig;
    setGenerationConfig(config: ImageGenerationConfig): void;
}
export interface InpaintingPipeline {
    new (): InpaintingPipeline;
    init(modelPath: string, device: string, properties: InpaintingPipelineProperties, callback: (err: Error | null) => void): void;
    generate(prompt: string, image: Tensor, mask: Tensor, properties: ImageGenerationConfig, streamer: ImageGenerationCallback | undefined, callback: (err: Error | null, result: Tensor) => void): void;
    decode(latent: Tensor, callback: (err: Error | null, result: Tensor) => void): void;
    getPerformanceMetrics(): ImageGenerationPerfMetrics;
    getGenerationConfig(): ImageGenerationConfig;
    setGenerationConfig(config: ImageGenerationConfig): void;
}
export interface Text2SpeechPipeline {
    new (): Text2SpeechPipeline;
    init(modelPath: string, device: string, properties: Text2SpeechPipelineProperties, callback: (err: Error | null) => void): void;
    generate(inputs: string | string[], speakerEmbedding: Tensor | undefined, properties: SpeechGenerationConfig, callback: (err: Error | null, result: {
        speeches: Tensor[];
        perfMetrics: Text2SpeechPerfMetrics;
    }) => void): void;
    getGenerationConfig(): SpeechGenerationConfig;
    setGenerationConfig(config: SpeechGenerationConfig): void;
}
export declare const TextEmbeddingPipeline: TextEmbeddingPipelineWrapper, TextRerankPipeline: TextRerankPipeline, LLMPipeline: LLMPipeline, VLMPipeline: VLMPipeline, WhisperPipeline: WhisperPipeline, Text2ImagePipeline: Text2ImagePipeline, Image2ImagePipeline: Image2ImagePipeline, InpaintingPipeline: InpaintingPipeline, Text2SpeechPipeline: Text2SpeechPipeline, ChatHistory: IChatHistory, Tokenizer: ITokenizer, ReasoningParser: IReasoningParser, DeepSeekR1ReasoningParser: IDeepSeekR1ReasoningParser, Phi4ReasoningParser: IPhi4ReasoningParser, Llama3PythonicToolParser: ILlama3PythonicToolParser, Llama3JsonToolParser: ILlama3JsonToolParser;
export type ChatHistory = IChatHistory;
export type Tokenizer = ITokenizer;
export type ReasoningParser = IReasoningParser;
export type DeepSeekR1ReasoningParser = IDeepSeekR1ReasoningParser;
export type Phi4ReasoningParser = IPhi4ReasoningParser;
export type Llama3PythonicToolParser = ILlama3PythonicToolParser;
export type Llama3JsonToolParser = ILlama3JsonToolParser;