Docs Query MCP

import type { FeatureExtractionPipeline, PipelineType } from "@xenova/transformers"; import { loadTransformersModule } from "./transformers-loader.js"; export class EmbeddingModel { private readonly modelName: string; private pipelinePromise?: Promise<FeatureExtractionPipeline>; constructor(modelName: string) { this.modelName = modelName; } private async loadPipeline(): Promise<FeatureExtractionPipeline> { if (!this.pipelinePromise) { this.pipelinePromise = (async () => { const { pipeline } = await loadTransformersModule(); return (pipeline("feature-extraction" as PipelineType, this.modelName, { quantized: true, }) as Promise<FeatureExtractionPipeline>); })(); } return this.pipelinePromise; } async embed(texts: string[]): Promise<number[][]> { if (!texts.length) { return []; } const extractor = await this.loadPipeline(); const embeddings: number[][] = []; for (const text of texts) { const result: any = await extractor(text, { pooling: "mean", normalize: true, }); const vector = Array.from(result.data as Float32Array).map((value) => Number(value)); embeddings.push(vector); } return embeddings; } }