🪨 feat: AWS Bedrock support (#3935)

* feat: Add BedrockIcon component to SVG library

* feat: EModelEndpoint.bedrock

* feat: first pass, bedrock chat. note: AgentClient is returning `agents` as conversation.endpoint

* fix: declare endpoint in initialization step

* chore: Update @librechat/agents dependency to version 1.4.5

* feat: backend content aggregation for agents/bedrock

* feat: abort agent requests

* feat: AWS Bedrock icons

* WIP: agent provider schema parsing

* chore: Update EditIcon props type

* refactor(useGenerationsByLatest): make agents and bedrock editable

* refactor: non-assistant message content, parts

* fix: Bedrock response `sender`

* fix: use endpointOption.model_parameters not endpointOption.modelOptions

* fix: types for step handler

* refactor: Update Agents.ToolCallDelta type

* refactor: Remove unnecessary assignment of parentMessageId in AskController

* refactor: remove unnecessary assignment of parentMessageId (agent request handler)

* fix(bedrock/agents): message regeneration

* refactor: dynamic form elements using react-hook-form Controllers

* fix: agent icons/labels for messages

* fix: agent actions

* fix: use of new dynamic tags causing application crash

* refactor: dynamic settings touch-ups

* refactor: update Slider component to allow custom track class name

* refactor: update DynamicSlider component styles

* refactor: use Constants value for GLOBAL_PROJECT_NAME (enum)

* feat: agent share global methods/controllers

* fix: agents query

* fix: `getResponseModel`

* fix: share prompt a11y issue

* refactor: update SharePrompt dialog theme styles

* refactor: explicit typing for SharePrompt

* feat: add agent roles/permissions

* chore: update @librechat/agents dependency to version 1.4.7 for tool_call_ids edge case

* fix(Anthropic): messages.X.content.Y.tool_use.input: Input should be a valid dictionary

* fix: handle text parts with tool_call_ids and empty text

* fix: role initialization

* refactor: don't make instructions required

* refactor: improve typing of Text part

* fix: setShowStopButton for agents route

* chore: remove params for now

* fix: add streamBuffer and streamRate to help prevent 'Overloaded' errors from Anthropic API

* refactor: remove console.log statement in ContentRender component

* chore: typing, rename Context to Delete Button

* chore(DeleteButton): logging

* refactor(Action): make accessible

* style(Action): improve a11y again

* refactor: remove use/mention of mongoose sessions

* feat: first pass, sharing agents

* feat: visual indicator for global agent, remove author when serving to non-author

* wip: params

* chore: fix typing issues

* fix(schemas): typing

* refactor: improve accessibility of ListCard component and fix console React warning

* wip: reset templates for non-legacy new convos

* Revert "wip: params"

This reverts commit f8067e91d4.

* Revert "refactor: dynamic form elements using react-hook-form Controllers"

This reverts commit 2150c4815d.

* fix(Parameters): types and parameter effect update to only update local state to parameters

* refactor: optimize useDebouncedInput hook for better performance

* feat: first pass, anthropic bedrock params

* chore: paramEndpoints check for endpointType too

* fix: maxTokens to use coerceNumber.optional(),

* feat: extra chat model params

* chore: reduce code repetition

* refactor: improve preset title handling in SaveAsPresetDialog component

* refactor: improve preset handling in HeaderOptions component

* chore: improve typing, replace legacy dialog for SaveAsPresetDialog

* feat: save as preset from parameters panel

* fix: multi-search in select dropdown when using Option type

* refactor: update default showDefault value to false in Dynamic components

* feat: Bedrock presets settings

* chore: config, fix agents schema, update config version

* refactor: update AWS region variable name in bedrock options endpoint to BEDROCK_AWS_DEFAULT_REGION

* refactor: update baseEndpointSchema in config.ts to include baseURL property

* refactor: update createRun function to include req parameter and set streamRate based on provider

* feat: availableRegions via config

* refactor: remove unused demo agent controller file

* WIP: title

* Update @librechat/agents to version 1.5.0

* chore: addTitle.js to handle empty responseText

* feat: support images and titles

* feat: context token updates

* Refactor BaseClient test to use expect.objectContaining

* refactor: add model select, remove header options params, move side panel params below prompts

* chore: update models list, catch title error

* feat: model service for bedrock models (env)

* chore: Remove verbose debug log in AgentClient class following stream

* feat(bedrock): track token spend; fix: token rates, value key mapping for AWS models

* refactor: handle streamRate in `handleLLMNewToken` callback

* chore: AWS Bedrock example config in `.env.example`

* refactor: Rename bedrockMeta to bedrockGeneral in settings.ts and use for AI21 and Amazon Bedrock providers

* refactor: Update `.env.example` with AWS Bedrock model IDs URL and additional notes

* feat: titleModel support for bedrock

* refactor: Update `.env.example` with additional notes for AWS Bedrock model IDs
This commit is contained in:
Danny Avila 2024-09-09 12:06:59 -04:00 committed by GitHub
parent 8c14360263
commit d59b62174f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
134 changed files with 3684 additions and 1213 deletions

View file

@ -165,7 +165,7 @@ async function createActionTool({ action, requestBuilder, zodSchema, name, descr
* Encrypts sensitive metadata values for an action.
*
* @param {ActionMetadata} metadata - The action metadata to encrypt.
* @returns {ActionMetadata} The updated action metadata with encrypted values.
* @returns {Promise<ActionMetadata>} The updated action metadata with encrypted values.
*/
async function encryptMetadata(metadata) {
const encryptedMetadata = { ...metadata };

View file

@ -94,18 +94,19 @@ const AppService = async (app) => {
);
}
if (endpoints?.[EModelEndpoint.openAI]) {
endpointLocals[EModelEndpoint.openAI] = endpoints[EModelEndpoint.openAI];
}
if (endpoints?.[EModelEndpoint.google]) {
endpointLocals[EModelEndpoint.google] = endpoints[EModelEndpoint.google];
}
if (endpoints?.[EModelEndpoint.anthropic]) {
endpointLocals[EModelEndpoint.anthropic] = endpoints[EModelEndpoint.anthropic];
}
if (endpoints?.[EModelEndpoint.gptPlugins]) {
endpointLocals[EModelEndpoint.gptPlugins] = endpoints[EModelEndpoint.gptPlugins];
}
const endpointKeys = [
EModelEndpoint.openAI,
EModelEndpoint.google,
EModelEndpoint.bedrock,
EModelEndpoint.anthropic,
EModelEndpoint.gptPlugins,
];
endpointKeys.forEach((key) => {
if (endpoints?.[key]) {
endpointLocals[key] = endpoints[key];
}
});
app.locals = {
...defaultLocals,

View file

@ -45,6 +45,7 @@ module.exports = {
AZURE_ASSISTANTS_BASE_URL,
EModelEndpoint.azureAssistants,
),
[EModelEndpoint.bedrock]: generateConfig(process.env.BEDROCK_AWS_SECRET_ACCESS_KEY),
/* key will be part of separate config */
[EModelEndpoint.agents]: generateConfig(process.env.I_AM_A_TEAPOT),
},

View file

@ -9,22 +9,13 @@ const { config } = require('./EndpointService');
*/
async function loadDefaultEndpointsConfig(req) {
const { google, gptPlugins } = await loadAsyncEndpoints(req);
const {
openAI,
agents,
assistants,
azureAssistants,
bingAI,
anthropic,
azureOpenAI,
chatGPTBrowser,
} = config;
const { assistants, azureAssistants, bingAI, azureOpenAI, chatGPTBrowser } = config;
const enabledEndpoints = getEnabledEndpoints();
const endpointConfig = {
[EModelEndpoint.openAI]: openAI,
[EModelEndpoint.agents]: agents,
[EModelEndpoint.openAI]: config[EModelEndpoint.openAI],
[EModelEndpoint.agents]: config[EModelEndpoint.agents],
[EModelEndpoint.assistants]: assistants,
[EModelEndpoint.azureAssistants]: azureAssistants,
[EModelEndpoint.azureOpenAI]: azureOpenAI,
@ -32,7 +23,8 @@ async function loadDefaultEndpointsConfig(req) {
[EModelEndpoint.bingAI]: bingAI,
[EModelEndpoint.chatGPTBrowser]: chatGPTBrowser,
[EModelEndpoint.gptPlugins]: gptPlugins,
[EModelEndpoint.anthropic]: anthropic,
[EModelEndpoint.anthropic]: config[EModelEndpoint.anthropic],
[EModelEndpoint.bedrock]: config[EModelEndpoint.bedrock],
};
const orderedAndFilteredEndpoints = enabledEndpoints.reduce((config, key, index) => {

View file

@ -3,6 +3,7 @@ const { useAzurePlugins } = require('~/server/services/Config/EndpointService').
const {
getOpenAIModels,
getGoogleModels,
getBedrockModels,
getAnthropicModels,
getChatGPTBrowserModels,
} = require('~/server/services/ModelService');
@ -38,6 +39,7 @@ async function loadDefaultModels(req) {
[EModelEndpoint.chatGPTBrowser]: chatGPTBrowser,
[EModelEndpoint.assistants]: assistants,
[EModelEndpoint.azureAssistants]: azureAssistants,
[EModelEndpoint.bedrock]: getBedrockModels(),
};
}

View file

@ -2,7 +2,7 @@ const { getAgent } = require('~/models/Agent');
const { logger } = require('~/config');
const buildOptions = (req, endpoint, parsedBody) => {
const { agent_id, instructions, spec, ...rest } = parsedBody;
const { agent_id, instructions, spec, ...model_parameters } = parsedBody;
const agentPromise = getAgent({
id: agent_id,
@ -19,9 +19,7 @@ const buildOptions = (req, endpoint, parsedBody) => {
agent_id,
instructions,
spec,
modelOptions: {
...rest,
},
model_parameters,
};
return endpointOption;

View file

@ -11,7 +11,12 @@
const { z } = require('zod');
const { tool } = require('@langchain/core/tools');
const { EModelEndpoint, providerEndpointMap } = require('librechat-data-provider');
const { createContentAggregator } = require('@librechat/agents');
const {
EModelEndpoint,
providerEndpointMap,
getResponseSender,
} = require('librechat-data-provider');
const { getDefaultHandlers } = require('~/server/controllers/agents/callbacks');
// for testing purposes
// const createTavilySearchTool = require('~/app/clients/tools/structured/TavilySearch');
@ -53,7 +58,8 @@ const initializeClient = async ({ req, res, endpointOption }) => {
}
// TODO: use endpointOption to determine options/modelOptions
const eventHandlers = getDefaultHandlers({ res });
const { contentParts, aggregateContent } = createContentAggregator();
const eventHandlers = getDefaultHandlers({ res, aggregateContent });
// const tools = [createTavilySearchTool()];
// const tools = [_getWeather];
@ -90,7 +96,7 @@ const initializeClient = async ({ req, res, endpointOption }) => {
}
// TODO: pass-in override settings that are specific to current run
endpointOption.modelOptions.model = agent.model;
endpointOption.model_parameters.model = agent.model;
const options = await getOptions({
req,
res,
@ -101,13 +107,21 @@ const initializeClient = async ({ req, res, endpointOption }) => {
});
modelOptions = Object.assign(modelOptions, options.llmConfig);
const sender = getResponseSender({
...endpointOption,
model: endpointOption.model_parameters.model,
});
const client = new AgentClient({
req,
agent,
tools,
sender,
toolMap,
contentParts,
modelOptions,
eventHandlers,
endpoint: EModelEndpoint.agents,
configOptions: options.configOptions,
maxContextTokens:
agent.max_context_tokens ??

View file

@ -23,7 +23,7 @@ const addTitle = async (req, { text, response, client }) => {
const title = await client.titleConvo({
text,
responseText: response?.text,
responseText: response?.text ?? '',
conversationId: response.conversationId,
});
await titleCache.set(key, title, 120000);

View file

@ -0,0 +1,44 @@
const { removeNullishValues, bedrockInputParser } = require('librechat-data-provider');
const generateArtifactsPrompt = require('~/app/clients/prompts/artifacts');
const { logger } = require('~/config');
const buildOptions = (endpoint, parsedBody) => {
const {
modelLabel: name,
promptPrefix,
maxContextTokens,
resendFiles = true,
imageDetail,
iconURL,
greeting,
spec,
artifacts,
...model_parameters
} = parsedBody;
let parsedParams = model_parameters;
try {
parsedParams = bedrockInputParser.parse(model_parameters);
} catch (error) {
logger.warn('Failed to parse bedrock input', error);
}
const endpointOption = removeNullishValues({
endpoint,
name,
resendFiles,
imageDetail,
iconURL,
greeting,
spec,
promptPrefix,
maxContextTokens,
model_parameters: parsedParams,
});
if (typeof artifacts === 'string') {
endpointOption.artifactsPrompt = generateArtifactsPrompt({ endpoint, artifacts });
}
return endpointOption;
};
module.exports = { buildOptions };

View file

@ -0,0 +1,7 @@
const build = require('./build');
const initialize = require('./initialize');
module.exports = {
...build,
...initialize,
};

View file

@ -0,0 +1,72 @@
const { createContentAggregator } = require('@librechat/agents');
const {
EModelEndpoint,
providerEndpointMap,
getResponseSender,
} = require('librechat-data-provider');
const { getDefaultHandlers } = require('~/server/controllers/agents/callbacks');
// const { loadAgentTools } = require('~/server/services/ToolService');
const getOptions = require('~/server/services/Endpoints/bedrock/options');
const AgentClient = require('~/server/controllers/agents/client');
const { getModelMaxTokens } = require('~/utils');
const initializeClient = async ({ req, res, endpointOption }) => {
if (!endpointOption) {
throw new Error('Endpoint option not provided');
}
/** @type {Array<UsageMetadata>} */
const collectedUsage = [];
const { contentParts, aggregateContent } = createContentAggregator();
const eventHandlers = getDefaultHandlers({ res, aggregateContent, collectedUsage });
// const tools = [createTavilySearchTool()];
/** @type {Agent} */
const agent = {
id: EModelEndpoint.bedrock,
name: endpointOption.name,
instructions: endpointOption.promptPrefix,
provider: EModelEndpoint.bedrock,
model: endpointOption.model_parameters.model,
model_parameters: endpointOption.model_parameters,
};
let modelOptions = { model: agent.model };
// TODO: pass-in override settings that are specific to current run
const options = await getOptions({
req,
res,
endpointOption,
});
modelOptions = Object.assign(modelOptions, options.llmConfig);
const maxContextTokens =
agent.max_context_tokens ??
getModelMaxTokens(modelOptions.model, providerEndpointMap[agent.provider]);
const sender = getResponseSender({
...endpointOption,
model: endpointOption.model_parameters.model,
});
const client = new AgentClient({
req,
agent,
sender,
// tools,
// toolMap,
modelOptions,
contentParts,
eventHandlers,
collectedUsage,
maxContextTokens,
endpoint: EModelEndpoint.bedrock,
configOptions: options.configOptions,
attachments: endpointOption.attachments,
});
return { client };
};
module.exports = { initializeClient };

View file

@ -0,0 +1,90 @@
const { HttpsProxyAgent } = require('https-proxy-agent');
const {
EModelEndpoint,
Constants,
AuthType,
removeNullishValues,
} = require('librechat-data-provider');
const { getUserKey, checkUserKeyExpiry } = require('~/server/services/UserService');
const { sleep } = require('~/server/utils');
const getOptions = async ({ req, endpointOption }) => {
const {
BEDROCK_AWS_SECRET_ACCESS_KEY,
BEDROCK_AWS_ACCESS_KEY_ID,
BEDROCK_REVERSE_PROXY,
BEDROCK_AWS_DEFAULT_REGION,
PROXY,
} = process.env;
const expiresAt = req.body.key;
const isUserProvided = BEDROCK_AWS_SECRET_ACCESS_KEY === AuthType.USER_PROVIDED;
const credentials = isUserProvided
? await getUserKey({ userId: req.user.id, name: EModelEndpoint.bedrock })
: {
accessKeyId: BEDROCK_AWS_ACCESS_KEY_ID,
secretAccessKey: BEDROCK_AWS_SECRET_ACCESS_KEY,
};
if (!credentials) {
throw new Error('Bedrock credentials not provided. Please provide them again.');
}
if (expiresAt && isUserProvided) {
checkUserKeyExpiry(expiresAt, EModelEndpoint.bedrock);
}
/** @type {number} */
let streamRate = Constants.DEFAULT_STREAM_RATE;
/** @type {undefined | TBaseEndpoint} */
const bedrockConfig = req.app.locals[EModelEndpoint.bedrock];
if (bedrockConfig && bedrockConfig.streamRate) {
streamRate = bedrockConfig.streamRate;
}
/** @type {undefined | TBaseEndpoint} */
const allConfig = req.app.locals.all;
if (allConfig && allConfig.streamRate) {
streamRate = allConfig.streamRate;
}
/** @type {import('@librechat/agents').BedrockConverseClientOptions} */
const requestOptions = Object.assign(
{
credentials,
model: endpointOption.model,
region: BEDROCK_AWS_DEFAULT_REGION,
streaming: true,
streamUsage: true,
callbacks: [
{
handleLLMNewToken: async () => {
if (!streamRate) {
return;
}
await sleep(streamRate);
},
},
],
},
endpointOption.model_parameters,
);
const configOptions = {};
if (PROXY) {
configOptions.httpAgent = new HttpsProxyAgent(PROXY);
}
if (BEDROCK_REVERSE_PROXY) {
configOptions.endpointHost = BEDROCK_REVERSE_PROXY;
}
return {
llmConfig: removeNullishValues(requestOptions),
configOptions,
};
};
module.exports = getOptions;

View file

@ -0,0 +1,40 @@
const { CacheKeys } = require('librechat-data-provider');
const getLogStores = require('~/cache/getLogStores');
const { isEnabled } = require('~/server/utils');
const { saveConvo } = require('~/models');
const addTitle = async (req, { text, response, client }) => {
const { TITLE_CONVO = true } = process.env ?? {};
if (!isEnabled(TITLE_CONVO)) {
return;
}
if (client.options.titleConvo === false) {
return;
}
// If the request was aborted, don't generate the title.
if (client.abortController.signal.aborted) {
return;
}
const titleCache = getLogStores(CacheKeys.GEN_TITLE);
const key = `${req.user.id}-${response.conversationId}`;
const title = await client.titleConvo({
text,
responseText: response?.text ?? '',
conversationId: response.conversationId,
});
await titleCache.set(key, title, 120000);
await saveConvo(
req,
{
conversationId: response.conversationId,
title,
},
{ context: 'api/server/services/Endpoints/bedrock/title.js' },
);
};
module.exports = addTitle;

View file

@ -49,7 +49,7 @@ const addTitle = async (req, { text, response, client }) => {
const title = await titleClient.titleConvo({
text,
responseText: response?.text,
responseText: response?.text ?? '',
conversationId: response.conversationId,
});
await titleCache.set(key, title, 120000);

View file

@ -23,7 +23,7 @@ const addTitle = async (req, { text, response, client }) => {
const title = await client.titleConvo({
text,
responseText: response?.text,
responseText: response?.text ?? '',
conversationId: response.conversationId,
});
await titleCache.set(key, title, 120000);

View file

@ -23,7 +23,13 @@ async function fetchImageToBase64(url) {
}
}
const base64Only = new Set([EModelEndpoint.google, EModelEndpoint.anthropic, 'Ollama', 'ollama']);
const base64Only = new Set([
EModelEndpoint.google,
EModelEndpoint.anthropic,
'Ollama',
'ollama',
EModelEndpoint.bedrock,
]);
/**
* Encodes and formats the given files.

View file

@ -5,6 +5,21 @@ const { extractBaseURL, inputSchema, processModelData, logAxiosError } = require
const { OllamaClient } = require('~/app/clients/OllamaClient');
const getLogStores = require('~/cache/getLogStores');
/**
* Splits a string by commas and trims each resulting value.
* @param {string} input - The input string to split.
* @returns {string[]} An array of trimmed values.
*/
const splitAndTrim = (input) => {
if (!input || typeof input !== 'string') {
return [];
}
return input
.split(',')
.map((item) => item.trim())
.filter(Boolean);
};
const { openAIApiKey, userProvidedOpenAI } = require('./Config/EndpointService').config;
/**
@ -194,7 +209,7 @@ const getOpenAIModels = async (opts) => {
}
if (process.env[key]) {
models = String(process.env[key]).split(',');
models = splitAndTrim(process.env[key]);
return models;
}
@ -208,7 +223,7 @@ const getOpenAIModels = async (opts) => {
const getChatGPTBrowserModels = () => {
let models = ['text-davinci-002-render-sha', 'gpt-4'];
if (process.env.CHATGPT_MODELS) {
models = String(process.env.CHATGPT_MODELS).split(',');
models = splitAndTrim(process.env.CHATGPT_MODELS);
}
return models;
@ -217,7 +232,7 @@ const getChatGPTBrowserModels = () => {
const getAnthropicModels = () => {
let models = defaultModels[EModelEndpoint.anthropic];
if (process.env.ANTHROPIC_MODELS) {
models = String(process.env.ANTHROPIC_MODELS).split(',');
models = splitAndTrim(process.env.ANTHROPIC_MODELS);
}
return models;
@ -226,7 +241,16 @@ const getAnthropicModels = () => {
const getGoogleModels = () => {
let models = defaultModels[EModelEndpoint.google];
if (process.env.GOOGLE_MODELS) {
models = String(process.env.GOOGLE_MODELS).split(',');
models = splitAndTrim(process.env.GOOGLE_MODELS);
}
return models;
};
const getBedrockModels = () => {
let models = defaultModels[EModelEndpoint.bedrock];
if (process.env.BEDROCK_AWS_MODELS) {
models = splitAndTrim(process.env.BEDROCK_AWS_MODELS);
}
return models;
@ -234,7 +258,9 @@ const getGoogleModels = () => {
module.exports = {
fetchModels,
splitAndTrim,
getOpenAIModels,
getBedrockModels,
getChatGPTBrowserModels,
getAnthropicModels,
getGoogleModels,

View file

@ -1,7 +1,16 @@
const axios = require('axios');
const { EModelEndpoint, defaultModels } = require('librechat-data-provider');
const { logger } = require('~/config');
const { fetchModels, getOpenAIModels } = require('./ModelService');
const {
fetchModels,
splitAndTrim,
getOpenAIModels,
getGoogleModels,
getBedrockModels,
getAnthropicModels,
} = require('./ModelService');
jest.mock('~/utils', () => {
const originalUtils = jest.requireActual('~/utils');
return {
@ -329,3 +338,71 @@ describe('fetchModels with Ollama specific logic', () => {
);
});
});
describe('splitAndTrim', () => {
it('should split a string by commas and trim each value', () => {
const input = ' model1, model2 , model3,model4 ';
const expected = ['model1', 'model2', 'model3', 'model4'];
expect(splitAndTrim(input)).toEqual(expected);
});
it('should return an empty array for empty input', () => {
expect(splitAndTrim('')).toEqual([]);
});
it('should return an empty array for null input', () => {
expect(splitAndTrim(null)).toEqual([]);
});
it('should return an empty array for undefined input', () => {
expect(splitAndTrim(undefined)).toEqual([]);
});
it('should filter out empty values after trimming', () => {
const input = 'model1,, ,model2,';
const expected = ['model1', 'model2'];
expect(splitAndTrim(input)).toEqual(expected);
});
});
describe('getAnthropicModels', () => {
it('returns default models when ANTHROPIC_MODELS is not set', () => {
delete process.env.ANTHROPIC_MODELS;
const models = getAnthropicModels();
expect(models).toEqual(defaultModels[EModelEndpoint.anthropic]);
});
it('returns models from ANTHROPIC_MODELS when set', () => {
process.env.ANTHROPIC_MODELS = 'claude-1, claude-2 ';
const models = getAnthropicModels();
expect(models).toEqual(['claude-1', 'claude-2']);
});
});
describe('getGoogleModels', () => {
it('returns default models when GOOGLE_MODELS is not set', () => {
delete process.env.GOOGLE_MODELS;
const models = getGoogleModels();
expect(models).toEqual(defaultModels[EModelEndpoint.google]);
});
it('returns models from GOOGLE_MODELS when set', () => {
process.env.GOOGLE_MODELS = 'gemini-pro, bard ';
const models = getGoogleModels();
expect(models).toEqual(['gemini-pro', 'bard']);
});
});
describe('getBedrockModels', () => {
it('returns default models when BEDROCK_AWS_MODELS is not set', () => {
delete process.env.BEDROCK_AWS_MODELS;
const models = getBedrockModels();
expect(models).toEqual(defaultModels[EModelEndpoint.bedrock]);
});
it('returns models from BEDROCK_AWS_MODELS when set', () => {
process.env.BEDROCK_AWS_MODELS = 'anthropic.claude-v2, ai21.j2-ultra ';
const models = getBedrockModels();
expect(models).toEqual(['anthropic.claude-v2', 'ai21.j2-ultra']);
});
});