Blame Raw
Cohee · e3f41666 · · 62 lines (2.2 KB)
1 contributor
1import fetch from 'node-fetch';
2import { setAdditionalHeadersByType } from '../additional-headers.js';
3import { TEXTGEN_TYPES } from '../constants.js';
4
5/**
6 * Gets the vector for the given text from Ollama
7 * @param {string[]} texts - The array of texts to get the vectors for
8 * @param {string} apiUrl - The API URL
9 * @param {string} model - The model to use
10 * @param {boolean} keep - Keep the model loaded in memory
11 * @param {import('../users.js').UserDirectoryList} directories - The directories object for the user
12 * @returns {Promise<number[][]>} - The array of vectors for the texts
13 */
14export async function getOllamaBatchVector(texts, apiUrl, model, keep, directories) {
15 const url = new URL(apiUrl);
16 url.pathname = '/api/embed';
17
18 const headers = {};
19 setAdditionalHeadersByType(headers, TEXTGEN_TYPES.OLLAMA, apiUrl, directories);
20
21 const response = await fetch(url, {
22 method: 'POST',
23 headers: {
24 'Content-Type': 'application/json',
25 ...headers,
26 },
27 body: JSON.stringify({
28 input: texts,
29 model: model,
30 keep_alive: keep ? -1 : undefined,
31 truncate: true,
32 }),
33 });
34
35 if (!response.ok) {
36 const responseText = await response.text();
37 throw new Error(`Ollama: Failed to get batch vectors: ${response.statusText} ${responseText}`);
38 }
39
40 /** @type {any} */
41 const data = await response.json();
42
43 if (!Array.isArray(data?.embeddings)) {
44 throw new Error('API response was not an array');
45 }
46
47 return data.embeddings;
48}
49
50/**
51 * Gets the vector for the given text from Ollama
52 * @param {string} text - The text to get the vector for
53 * @param {string} apiUrl - The API URL
54 * @param {string} model - The model to use
55 * @param {boolean} keep - Keep the model loaded in memory
56 * @param {import('../users.js').UserDirectoryList} directories - The directories object for the user
57 * @returns {Promise<number[]>} - The vector for the text
58 */
59export async function getOllamaVector(text, apiUrl, model, keep, directories) {
60 const vectors = await getOllamaBatchVector([text], apiUrl, model, keep, directories);
61 return vectors[0];
62}