rixty/SillyTavern
1
0
Fork 0
mirror of https://github.com/SillyTavern/SillyTavern.git synced 2025-06-05 21:59:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'staging' into continue-from-reasoning

Browse Source
This commit is contained in:
Cohee
2025-03-09 01:19:25 +02:00
parent 980ed76cc3 b52b11d7bb
commit 96d79ac4e9
13 changed files with 303 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
public/index.html
View File

@@ -1962,7 +1962,7 @@
</span>
</div>
</div>
<div class="range-block" data-source="openai,cohere,mistralai,custom,claude,openrouter,groq,deepseek,makersuite">
<div class="range-block" data-source="openai,cohere,mistralai,custom,claude,openrouter,groq,deepseek,makersuite,ai21">
<label for="openai_function_calling" class="checkbox_label flexWrap widthFreeExpand">
<input id="openai_function_calling" type="checkbox" />
<span data-i18n="Enable function calling">Enable function calling</span>
@@ -3078,7 +3078,15 @@
<div>
<h4 data-i18n="AI21 Model">AI21 Model</h4>
<select id="model_ai21_select">
<optgroup label="Jamba 1.5">
<optgroup label="Jamba (Latest)">
<option value="jamba-mini">jamba-mini</option>
<option value="jamba-large">jamba-large</option>
</optgroup>
<optgroup label="Jamba 1.6">
<option value="jamba-1.6-mini">jamba-1.6-mini</option>
<option value="jamba-1.6-large">jamba-1.6-large</option>
</optgroup>
<optgroup label="Jamba 1.5 (Deprecated)">
<option value="jamba-1.5-mini">jamba-1.5-mini</option>
<option value="jamba-1.5-large">jamba-1.5-large</option>
</optgroup>

25
public/script.js
View File

@@ -271,7 +271,7 @@ import { initSettingsSearch } from './scripts/setting-search.js';
import { initBulkEdit } from './scripts/bulk-edit.js';
import { deriveTemplatesFromChatTemplate } from './scripts/chat-templates.js';
import { getContext } from './scripts/st-context.js';
import { extractReasoningFromData, initReasoning, PromptReasoning, ReasoningHandler, removeReasoningFromString, updateReasoningUI } from './scripts/reasoning.js';
import { extractReasoningFromData, initReasoning, parseReasoningInSwipes, PromptReasoning, ReasoningHandler, removeReasoningFromString, updateReasoningUI } from './scripts/reasoning.js';
import { accountStorage } from './scripts/util/AccountStorage.js';
// API OBJECT FOR EXTERNAL WIRING
@@ -3346,15 +3346,18 @@ class StreamingProcessor {
if (Array.isArray(this.swipes) && this.swipes.length > 0) {
const message = chat[messageId];
const swipeInfoExtra = structuredClone(message.extra ?? {});
delete swipeInfoExtra.token_count;
delete swipeInfoExtra.reasoning;
delete swipeInfoExtra.reasoning_duration;
const swipeInfo = {
send_date: message.send_date,
gen_started: message.gen_started,
gen_finished: message.gen_finished,
extra: structuredClone(message.extra),
extra: swipeInfoExtra,
};
const swipeInfoArray = [];
swipeInfoArray.length = this.swipes.length;
swipeInfoArray.fill(swipeInfo);
const swipeInfoArray = Array(this.swipes.length).fill().map(() => structuredClone(swipeInfo));
parseReasoningInSwipes(this.swipes, swipeInfoArray, message.extra?.reasoning_duration);
chat[messageId].swipes.push(...this.swipes);
chat[messageId].swipe_info.push(...swipeInfoArray);
}
@@ -3366,6 +3369,7 @@ class StreamingProcessor {
await eventSource.emit(event_types.IMPERSONATE_READY, text);
}
syncMesToSwipe(messageId);
saveLogprobsForActiveMessage(this.messageLogprobs.filter(Boolean), this.continueMessage);
await saveChatConditional();
unblockGeneration();
@@ -6117,15 +6121,18 @@ export async function saveReply(type, getMessage, fromStreaming, title, swipes,
}
if (Array.isArray(swipes) && swipes.length > 0) {
const swipeInfoExtra = structuredClone(item.extra ?? {});
delete swipeInfoExtra.token_count;
delete swipeInfoExtra.reasoning;
delete swipeInfoExtra.reasoning_duration;
const swipeInfo = {
send_date: item.send_date,
gen_started: item.gen_started,
gen_finished: item.gen_finished,
extra: structuredClone(item.extra),
extra: swipeInfoExtra,
};
const swipeInfoArray = [];
swipeInfoArray.length = swipes.length;
swipeInfoArray.fill(swipeInfo, 0, swipes.length);
const swipeInfoArray = Array(swipes.length).fill().map(() => structuredClone(swipeInfo));
parseReasoningInSwipes(swipes, swipeInfoArray, item.extra?.reasoning_duration);
item.swipes.push(...swipes);
item.swipe_info.push(...swipeInfoArray);
}

2
public/scripts/extensions.js
View File

@@ -1070,7 +1070,7 @@ export async function installExtension(url, global) {
toastr.success(t`Extension '${response.display_name}' by ${response.author} (version ${response.version}) has been installed successfully!`, t`Extension installation successful`);
console.debug(`Extension "${response.display_name}" has been installed successfully at ${response.extensionPath}`);
await loadExtensionSettings({}, false, false);
await eventSource.emit(event_types.EXTENSION_SETTINGS_LOADED);
await eventSource.emit(event_types.EXTENSION_SETTINGS_LOADED, response);
}
/**

144
public/scripts/extensions/vectors/index.js
View File

@@ -19,6 +19,7 @@ import {
modules,
renderExtensionTemplateAsync,
doExtrasFetch, getApiUrl,
openThirdPartyExtensionMenu,
} from '../../extensions.js';
import { collapseNewlines, registerDebugFunction } from '../../power-user.js';
import { SECRET_KEYS, secret_state, writeSecret } from '../../secrets.js';
@@ -34,6 +35,7 @@ import { SlashCommandEnumValue, enumTypes } from '../../slash-commands/SlashComm
import { slashCommandReturnHelper } from '../../slash-commands/SlashCommandReturnHelper.js';
import { callGenericPopup, POPUP_RESULT, POPUP_TYPE } from '../../popup.js';
import { generateWebLlmChatPrompt, isWebLlmSupported } from '../shared.js';
import { WebLlmVectorProvider } from './webllm.js';
/**
* @typedef {object} HashedMessage
@@ -60,6 +62,7 @@ const settings = {
ollama_model: 'mxbai-embed-large',
ollama_keep: false,
vllm_model: '',
webllm_model: '',
summarize: false,
summarize_sent: false,
summary_source: 'main',
@@ -103,7 +106,7 @@ const settings = {
};
const moduleWorker = new ModuleWorkerWrapper(synchronizeChat);
const webllmProvider = new WebLlmVectorProvider();
const cachedSummaries = new Map();
/**
@@ -373,6 +376,8 @@ async function synchronizeChat(batchSize = 5) {
return 'Vectorization Source Model is required, but not set.';
case 'extras_module_missing':
return 'Extras API must provide an "embeddings" module.';
case 'webllm_not_supported':
return 'WebLLM extension is not installed or the model is not set.';
default:
return 'Check server console for more details';
}
@@ -747,14 +752,15 @@ async function getQueryText(chat, initiator) {
/**
* Gets common body parameters for vector requests.
* @returns {object}
* @param {object} args Additional arguments
* @returns {object} Request body
*/
function getVectorsRequestBody() {
const body = {};
function getVectorsRequestBody(args = {}) {
const body = Object.assign({}, args);
switch (settings.source) {
case 'extras':
body.extrasUrl = extension_settings.apiUrl;
body.extrasKey = extension_settings.apiKey;
body.extrasUrl = extension_settings.apiUrl;
body.extrasKey = extension_settings.apiKey;
break;
case 'togetherai':
body.model = extension_settings.vectors.togetherai_model;
@@ -777,12 +783,30 @@ function getVectorsRequestBody() {
body.apiUrl = textgenerationwebui_settings.server_urls[textgen_types.VLLM];
body.model = extension_settings.vectors.vllm_model;
break;
case 'webllm':
body.model = extension_settings.vectors.webllm_model;
break;
default:
break;
}
return body;
}
/**
* Gets additional arguments for vector requests.
* @param {string[]} items Items to embed
* @returns {Promise<object>} Additional arguments
*/
async function getAdditionalArgs(items) {
const args = {};
switch (settings.source) {
case 'webllm':
args.embeddings = await createWebLlmEmbeddings(items);
break;
}
return args;
}
/**
* Gets the saved hashes for a collection
* @param {string} collectionId
@@ -816,11 +840,12 @@ async function getSavedHashes(collectionId) {
async function insertVectorItems(collectionId, items) {
throwIfSourceInvalid();
const args = await getAdditionalArgs(items.map(x => x.text));
const response = await fetch('/api/vector/insert', {
method: 'POST',
headers: getRequestHeaders(),
body: JSON.stringify({
...getVectorsRequestBody(),
...getVectorsRequestBody(args),
collectionId: collectionId,
items: items,
source: settings.source,
@@ -858,6 +883,10 @@ function throwIfSourceInvalid() {
if (settings.source === 'extras' && !modules.includes('embeddings')) {
throw new Error('Vectors: Embeddings module missing', { cause: 'extras_module_missing' });
}
if (settings.source === 'webllm' && (!isWebLlmSupported() || !settings.webllm_model)) {
throw new Error('Vectors: WebLLM is not supported', { cause: 'webllm_not_supported' });
}
}
/**
@@ -890,11 +919,12 @@ async function deleteVectorItems(collectionId, hashes) {
* @returns {Promise<{ hashes: number[], metadata: object[]}>} - Hashes of the results
*/
async function queryCollection(collectionId, searchText, topK) {
const args = await getAdditionalArgs([searchText]);
const response = await fetch('/api/vector/query', {
method: 'POST',
headers: getRequestHeaders(),
body: JSON.stringify({
...getVectorsRequestBody(),
...getVectorsRequestBody(args),
collectionId: collectionId,
searchText: searchText,
topK: topK,
@@ -919,11 +949,12 @@ async function queryCollection(collectionId, searchText, topK) {
* @returns {Promise<Record<string, { hashes: number[], metadata: object[] }>>} - Results mapped to collection IDs
*/
async function queryMultipleCollections(collectionIds, searchText, topK, threshold) {
const args = await getAdditionalArgs([searchText]);
const response = await fetch('/api/vector/query-multi', {
method: 'POST',
headers: getRequestHeaders(),
body: JSON.stringify({
...getVectorsRequestBody(),
...getVectorsRequestBody(args),
collectionIds: collectionIds,
searchText: searchText,
topK: topK,
@@ -1039,6 +1070,72 @@ function toggleSettings() {
$('#llamacpp_vectorsModel').toggle(settings.source === 'llamacpp');
$('#vllm_vectorsModel').toggle(settings.source === 'vllm');
$('#nomicai_apiKey').toggle(settings.source === 'nomicai');
$('#webllm_vectorsModel').toggle(settings.source === 'webllm');
if (settings.source === 'webllm') {
loadWebLlmModels();
}
}
/**
* Executes a function with WebLLM error handling.
* @param {function(): Promise<T>} func Function to execute
* @returns {Promise<T>}
* @template T
*/
async function executeWithWebLlmErrorHandling(func) {
try {
return await func();
} catch (error) {
console.log('Vectors: Failed to load WebLLM models', error);
if (!(error instanceof Error)) {
return;
}
switch (error.cause) {
case 'webllm-not-available':
toastr.warning('WebLLM is not available. Please install the extension.', 'WebLLM not installed');
break;
case 'webllm-not-updated':
toastr.warning('The installed extension version does not support embeddings.', 'WebLLM update required');
break;
}
}
}
/**
* Loads and displays WebLLM models in the settings.
* @returns {Promise<void>}
*/
function loadWebLlmModels() {
return executeWithWebLlmErrorHandling(() => {
const models = webllmProvider.getModels();
$('#vectors_webllm_model').empty();
for (const model of models) {
$('#vectors_webllm_model').append($('<option>', { value: model.id, text: model.toString() }));
}
if (!settings.webllm_model || !models.some(x => x.id === settings.webllm_model)) {
if (models.length) {
settings.webllm_model = models[0].id;
}
}
$('#vectors_webllm_model').val(settings.webllm_model);
return Promise.resolve();
});
}
/**
* Creates WebLLM embeddings for a list of items.
* @param {string[]} items Items to embed
* @returns {Promise<Record<string, number[]>>} Calculated embeddings
*/
async function createWebLlmEmbeddings(items) {
return executeWithWebLlmErrorHandling(async () => {
const embeddings = await webllmProvider.embedTexts(items, settings.webllm_model);
const result = /** @type {Record<string, number[]>} */ ({});
for (let i = 0; i < items.length; i++) {
result[items[i]] = embeddings[i];
}
return result;
});
}
async function onPurgeClick() {
@@ -1567,6 +1664,30 @@ jQuery(async () => {
$('#dialogue_popup_input').val(presetModel);
});
$('#vectors_webllm_install').on('click', (e) => {
e.preventDefault();
e.stopPropagation();
if (Object.hasOwn(SillyTavern, 'llm')) {
toastr.info('WebLLM is already installed');
return;
}
openThirdPartyExtensionMenu('https://github.com/SillyTavern/Extension-WebLLM');
});
$('#vectors_webllm_model').on('input', () => {
settings.webllm_model = String($('#vectors_webllm_model').val());
Object.assign(extension_settings.vectors, settings);
saveSettingsDebounced();
});
$('#vectors_webllm_load').on('click', async () => {
if (!settings.webllm_model) return;
await webllmProvider.loadModel(settings.webllm_model);
toastr.success('WebLLM model loaded');
});
$('#api_key_nomicai').toggleClass('success', !!secret_state[SECRET_KEYS.NOMICAI]);
toggleSettings();
@@ -1578,6 +1699,11 @@ jQuery(async () => {
eventSource.on(event_types.CHAT_DELETED, purgeVectorIndex);
eventSource.on(event_types.GROUP_CHAT_DELETED, purgeVectorIndex);
eventSource.on(event_types.FILE_ATTACHMENT_DELETED, purgeFileVectorIndex);
eventSource.on(event_types.EXTENSION_SETTINGS_LOADED, async (manifest) => {
if (settings.source === 'webllm' && manifest?.display_name === 'WebLLM') {
await loadWebLlmModels();
}
});
SlashCommandParser.addCommandObject(SlashCommand.fromProps({
name: 'db-ingest',

16
public/scripts/extensions/vectors/settings.html
View File

@@ -21,8 +21,24 @@
<option value="openai">OpenAI</option>
<option value="togetherai">TogetherAI</option>
<option value="vllm">vLLM</option>
<option value="webllm" data-i18n="WebLLM Extension">WebLLM Extension</option>
</select>
</div>
<div class="flex-container flexFlowColumn" id="webllm_vectorsModel">
<label for="vectors_webllm_model" data-i18n="Vectorization Model">
Vectorization Model
</label>
<div class="flex-container">
<select id="vectors_webllm_model" class="text_pole flex1">
</select>
<div id="vectors_webllm_load" class="menu_button menu_button_icon" title="Verify and load the selected model.">
<i class="fa-solid fa-check-to-slot"></i>
</div>
</div>
<div>
Requires the WebLLM extension to be installed. Click <a href="#" id="vectors_webllm_install">here</a> to install.
</div>
</div>
<div class="flex-container flexFlowColumn" id="ollama_vectorsModel">
<label for="vectors_ollama_model" data-i18n="Vectorization Model">
Vectorization Model

64
public/scripts/extensions/vectors/webllm.js Normal file
View File

@@ -0,0 +1,64 @@
export class WebLlmVectorProvider {
/** @type {object?} WebLLM engine */
#engine = null;
constructor() {
this.#engine = null;
}
/**
* Check if WebLLM is available and up-to-date
* @throws {Error} If WebLLM is not available or not up-to-date
*/
#checkWebLlm() {
if (!Object.hasOwn(SillyTavern, 'llm')) {
throw new Error('WebLLM is not available', { cause: 'webllm-not-available' });
}
if (typeof SillyTavern.llm.generateEmbedding !== 'function') {
throw new Error('WebLLM is not updated', { cause: 'webllm-not-updated' });
}
}
/**
* Initialize the engine with a model.
* @param {string} modelId Model ID to initialize the engine with
* @returns {Promise<void>} Promise that resolves when the engine is initialized
*/
#initEngine(modelId) {
this.#checkWebLlm();
if (!this.#engine) {
this.#engine = SillyTavern.llm.getEngine();
}
return this.#engine.loadModel(modelId);
}
/**
* Get available models.
* @returns {{id:string, toString: function(): string}[]} Array of available models
*/
getModels() {
this.#checkWebLlm();
return SillyTavern.llm.getEmbeddingModels();
}
/**
* Generate embeddings for a list of texts.
* @param {string[]} texts Array of texts to generate embeddings for
* @param {string} modelId Model to use for generating embeddings
* @returns {Promise<number[][]>} Array of embeddings for each text
*/
async embedTexts(texts, modelId) {
await this.#initEngine(modelId);
return this.#engine.generateEmbedding(texts);
}
/**
* Loads a model into the engine.
* @param {string} modelId Model ID to load
*/
async loadModel(modelId) {
await this.#initEngine(modelId);
}
}

18
public/scripts/openai.js
View File

@@ -337,7 +337,7 @@ const default_settings = {
openai_model: 'gpt-4-turbo',
claude_model: 'claude-3-5-sonnet-20240620',
google_model: 'gemini-1.5-pro',
ai21_model: 'jamba-1.5-large',
ai21_model: 'jamba-1.6-large',
mistralai_model: 'mistral-large-latest',
cohere_model: 'command-r-plus',
perplexity_model: 'sonar-pro',
@@ -417,7 +417,7 @@ const oai_settings = {
openai_model: 'gpt-4-turbo',
claude_model: 'claude-3-5-sonnet-20240620',
google_model: 'gemini-1.5-pro',
ai21_model: 'jamba-1.5-large',
ai21_model: 'jamba-1.6-large',
mistralai_model: 'mistral-large-latest',
cohere_model: 'command-r-plus',
perplexity_model: 'sonar-pro',
@@ -2027,12 +2027,16 @@ async function sendOpenAIRequest(type, messages, signal) {
generate_data['logprobs'] = 5;
}
// Remove logit bias, logprobs and stop strings if it's not supported by the model
if (isOAI && oai_settings.openai_model.includes('vision') || isOpenRouter && oai_settings.openrouter_model.includes('vision') || isOAI && oai_settings.openai_model.includes('gpt-4.5-preview')) {
// Remove logit bias/logprobs/stop-strings if not supported by the model
const isVision = (m) => ['gpt', 'vision'].every(x => m.includes(x));
if (isOAI && isVision(oai_settings.openai_model) || isOpenRouter && isVision(oai_settings.openrouter_model)) {
delete generate_data.logit_bias;
delete generate_data.stop;
delete generate_data.logprobs;
}
if (isOAI && oai_settings.openai_model.includes('gpt-4.5-preview') || isOpenRouter && oai_settings.openrouter_model.includes('gpt-4.5-preview')) {
delete generate_data.logprobs;
}
if (isClaude) {
generate_data['top_k'] = Number(oai_settings.top_k_openai);
@@ -3251,7 +3255,7 @@ function loadOpenAISettings(data, settings) {
}
if (oai_settings.ai21_model.startsWith('j2-')) {
oai_settings.ai21_model = 'jamba-1.5-large';
oai_settings.ai21_model = 'jamba-1.6-large';
}
if (settings.wrap_in_quotes !== undefined) oai_settings.wrap_in_quotes = !!settings.wrap_in_quotes;
@@ -4208,7 +4212,7 @@ async function onModelChange() {
if ($(this).is('#model_ai21_select')) {
if (value === '' || value.startsWith('j2-')) {
value = 'jamba-1.5-large';
value = 'jamba-1.6-large';
$('#model_ai21_select').val(value);
}
@@ -4485,7 +4489,7 @@ async function onModelChange() {
if (oai_settings.chat_completion_source == chat_completion_sources.AI21) {
if (oai_settings.max_context_unlocked) {
$('#openai_max_context').attr('max', unlocked_max);
} else if (oai_settings.ai21_model.includes('jamba-1.5') || oai_settings.ai21_model.includes('jamba-instruct')) {
} else if (oai_settings.ai21_model.startsWith('jamba-')) {
$('#openai_max_context').attr('max', max_256k);
}

26
public/scripts/reasoning.js
View File

@@ -1104,6 +1104,32 @@ function parseReasoningFromString(str, { strict = true } = {}) {
}
}
/**
* Parse reasoning in an array of swipe strings if auto-parsing is enabled.
* @param {string[]} swipes Array of swipe strings
* @param {{extra: {reasoning: string, reasoning_duration: number}}[]} swipeInfoArray Array of swipe info objects
* @param {number?} duration Duration of the reasoning
*/
export function parseReasoningInSwipes(swipes, swipeInfoArray, duration) {
if (!power_user.reasoning.auto_parse) {
return;
}
// Something ain't right, don't parse
if (!Array.isArray(swipes) || !Array.isArray(swipeInfoArray) || swipes.length !== swipeInfoArray.length) {
return;
}
for (let index = 0; index < swipes.length; index++) {
const parsedReasoning = parseReasoningFromString(swipes[index]);
if (parsedReasoning) {
swipes[index] = parsedReasoning.content;
swipeInfoArray[index].extra.reasoning = parsedReasoning.reasoning;
swipeInfoArray[index].extra.reasoning_duration = duration;
}
}
}
function registerReasoningAppEvents() {
const eventHandler = (/** @type {string} */ type, /** @type {number} */ idx) => {
if (!power_user.reasoning.auto_parse) {

1
public/scripts/tool-calling.js
View File

@@ -585,6 +585,7 @@ export class ToolManager {
chat_completion_sources.COHERE,
chat_completion_sources.DEEPSEEK,
chat_completion_sources.MAKERSUITE,
chat_completion_sources.AI21,
];
return supportedSources.includes(oai_settings.chat_completion_source);
}

9
src/endpoints/backends/chat-completions.js
View File

@@ -499,6 +499,12 @@ async function sendMakerSuiteRequest(request, response) {
async function sendAI21Request(request, response) {
if (!request.body) return response.sendStatus(400);
const apiKey = readSecret(request.user.directories, SECRET_KEYS.AI21);
if (!apiKey) {
console.warn('AI21 API key is missing.');
return response.status(400).send({ error: true });
}
const controller = new AbortController();
console.debug(request.body.messages);
request.socket.removeAllListeners('close');
@@ -514,13 +520,14 @@ async function sendAI21Request(request, response) {
top_p: request.body.top_p,
stop: request.body.stop,
stream: request.body.stream,
tools: request.body.tools,
};
const options = {
method: 'POST',
headers: {
accept: 'application/json',
'content-type': 'application/json',
Authorization: `Bearer ${readSecret(request.user.directories, SECRET_KEYS.AI21)}`,
Authorization: `Bearer ${apiKey}`,
},
body: JSON.stringify(body),
signal: controller.signal,

2
src/endpoints/characters.js
View File

@@ -218,11 +218,13 @@ const toShallow = (character) => {
date_last_chat: character.date_last_chat,
chat_size: character.chat_size,
data_size: character.data_size,
tags: character.tags,
data: {
name: _.get(character, 'data.name', ''),
character_version: _.get(character, 'data.character_version', ''),
creator: _.get(character, 'data.creator', ''),
creator_notes: _.get(character, 'data.creator_notes', ''),
tags: _.get(character, 'data.tags', []),
extensions: {
fav: _.get(character, 'data.extensions.fav', false),
},

4
src/endpoints/users-admin.js
View File

@@ -4,7 +4,7 @@ import storage from 'node-persist';
import express from 'express';
import lodash from 'lodash';
import { jsonParser } from '../express-common.js';
import { checkForNewContent } from './content-manager.js';
import { checkForNewContent, CONTENT_TYPES } from './content-manager.js';
import {
KEY_PREFIX,
toKey,
@@ -195,7 +195,7 @@ router.post('/create', requireAdminMiddleware, jsonParser, async (request, respo
console.info('Creating data directories for', newUser.handle);
await ensurePublicDirectoriesExist();
const directories = getUserDirectories(newUser.handle);
await checkForNewContent([directories]);
await checkForNewContent([directories], [CONTENT_TYPES.SETTINGS]);
return response.json({ handle: newUser.handle });
} catch (error) {
console.error('User create failed:', error);

11
src/endpoints/vectors.js
View File

@@ -31,6 +31,7 @@ const SOURCES = [
'ollama',
'llamacpp',
'vllm',
'webllm',
];
/**
@@ -64,6 +65,8 @@ async function getVector(source, sourceSettings, text, isQuery, directories) {
return getVllmVector(text, sourceSettings.apiUrl, sourceSettings.model, directories);
case 'ollama':
return getOllamaVector(text, sourceSettings.apiUrl, sourceSettings.model, sourceSettings.keep, directories);
case 'webllm':
return sourceSettings.embeddings[text];
}
throw new Error(`Unknown vector source ${source}`);
@@ -114,6 +117,9 @@ async function getBatchVector(source, sourceSettings, texts, isQuery, directorie
case 'ollama':
results.push(...await getOllamaBatchVector(batch, sourceSettings.apiUrl, sourceSettings.model, sourceSettings.keep, directories));
break;
case 'webllm':
results.push(...texts.map(x => sourceSettings.embeddings[x]));
break;
default:
throw new Error(`Unknown vector source ${source}`);
}
@@ -179,6 +185,11 @@ function getSourceSettings(source, request) {
return {
model: 'nomic-embed-text-v1.5',
};
case 'webllm':
return {
model: String(request.body.model),
embeddings: request.body.embeddings ?? {},
};
default:
return {};
}