ztimson/ai-utils
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

One embedding at a time
All checks were successful
Publish Library / Build NPM Project (push) Successful in 27s
Details
Publish Library / Tag Version (push) Successful in 7s
Details

Browse Source
This commit is contained in:
Zakary Timson
2026-02-19 22:58:53 -05:00
parent da15d299e6
commit a07f069ad0
2 changed files with 10 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
package.json
View File

@@ -1,6 +1,6 @@
{ {
"name": "@ztimson/ai-utils", "name": "@ztimson/ai-utils",
"version": "0.7.2", "version": "0.7.3",
"description": "AI Utility library", "description": "AI Utility library",
"author": "Zak Timson", "author": "Zak Timson",
"license": "MIT", "license": "MIT",

25
src/llm.ts
View File

@@ -255,12 +255,11 @@ class LLM {
/** /**
* Create a vector representation of a string * Create a vector representation of a string
* @param {object | string} target Item that will be embedded (objects get converted) * @param {object | string} target Item that will be embedded (objects get converted)
* @param {maxTokens?: number, overlapTokens?: number, parellel?: number} opts Options for embedding such as chunk sizes and parallel processing * @param {maxTokens?: number, overlapTokens?: number} opts Options for embedding such as chunk sizes
* @returns {Promise<Awaited<{index: number, embedding: number[], text: string, tokens: number}>[]>} Chunked embeddings * @returns {Promise<Awaited<{index: number, embedding: number[], text: string, tokens: number}>[]>} Chunked embeddings
*/ */
async embedding(target: object | string, opts: {maxTokens?: number, overlapTokens?: number, parallel?: number} = {}) { async embedding(target: object | string, opts: {maxTokens?: number, overlapTokens?: number} = {}) {
let {maxTokens = 500, overlapTokens = 50, parallel = 1} = opts; let {maxTokens = 500, overlapTokens = 50} = opts;
const embed = (text: string): Promise<number[]> => { const embed = (text: string): Promise<number[]> => {
return new Promise((resolve, reject) => { return new Promise((resolve, reject) => {
const worker = new Worker(join(dirname(fileURLToPath(import.meta.url)), 'embedder.js')); const worker = new Worker(join(dirname(fileURLToPath(import.meta.url)), 'embedder.js'));
@@ -280,19 +279,13 @@ class LLM {
worker.postMessage({text, model: this.ai.options?.embedder || 'bge-small-en-v1.5', modelDir: this.ai.options.path}); worker.postMessage({text, model: this.ai.options?.embedder || 'bge-small-en-v1.5', modelDir: this.ai.options.path});
}); });
}; };
let i = 0, chunks = this.chunk(target, maxTokens, overlapTokens), results: any[] = []; const chunks = this.chunk(target, maxTokens, overlapTokens), results: any[] = [];
const next: Function = () => { for(let i = 0; i < chunks.length; i++) {
const index = i++; const text= chunks[i];
if(index >= chunks.length) return; const embedding = await embed(text);
const text = chunks[index]; results.push({index: i, embedding, text, tokens: this.estimateTokens(text)});
return embed(text).then(embedding => {
results.push({index, embedding, text, tokens: this.estimateTokens(text)});
return next();
})
} }
return results;
await Promise.all(Array(parallel).fill(null).map(() => next()));
return results.toSorted((a, b) => a.index - b.index);
} }
/** /**