|
| 1 | +import ciInfo from "ci-info"; |
| 2 | +import got from "got"; |
| 3 | +import ora from "ora"; |
| 4 | +import { red } from "picocolors"; |
| 5 | +import prompts from "prompts"; |
| 6 | +import { ModelConfigParams } from "."; |
| 7 | +import { questionHandlers } from "../../questions"; |
| 8 | + |
| 9 | +export const TSYSTEMS_LLMHUB_API_URL = |
| 10 | + "https://llm-server.llmhub.t-systems.net/v2"; |
| 11 | + |
| 12 | +const DEFAULT_MODEL = "gpt-3.5-turbo"; |
| 13 | +const DEFAULT_EMBEDDING_MODEL = "text-embedding-3-large"; |
| 14 | + |
| 15 | +const LLMHUB_MODELS = [ |
| 16 | + "gpt-35-turbo", |
| 17 | + "gpt-4-32k-1", |
| 18 | + "gpt-4-32k-canada", |
| 19 | + "gpt-4-32k-france", |
| 20 | + "gpt-4-turbo-128k-france", |
| 21 | + "Llama2-70b-Instruct", |
| 22 | + "Llama-3-70B-Instruct", |
| 23 | + "Mixtral-8x7B-Instruct-v0.1", |
| 24 | + "mistral-large-32k-france", |
| 25 | + "CodeLlama-2", |
| 26 | +]; |
| 27 | +const LLMHUB_EMBEDDING_MODELS = [ |
| 28 | + "text-embedding-ada-002", |
| 29 | + "text-embedding-ada-002-france", |
| 30 | + "jina-embeddings-v2-base-de", |
| 31 | + "jina-embeddings-v2-base-code", |
| 32 | + "text-embedding-bge-m3", |
| 33 | +]; |
| 34 | + |
| 35 | +type LLMHubQuestionsParams = { |
| 36 | + apiKey?: string; |
| 37 | + askModels: boolean; |
| 38 | +}; |
| 39 | + |
| 40 | +export async function askLLMHubQuestions({ |
| 41 | + askModels, |
| 42 | + apiKey, |
| 43 | +}: LLMHubQuestionsParams): Promise<ModelConfigParams> { |
| 44 | + const config: ModelConfigParams = { |
| 45 | + apiKey, |
| 46 | + model: DEFAULT_MODEL, |
| 47 | + embeddingModel: DEFAULT_EMBEDDING_MODEL, |
| 48 | + dimensions: getDimensions(DEFAULT_EMBEDDING_MODEL), |
| 49 | + isConfigured(): boolean { |
| 50 | + if (config.apiKey) { |
| 51 | + return true; |
| 52 | + } |
| 53 | + if (process.env["T_SYSTEMS_LLMHUB_API_KEY"]) { |
| 54 | + return true; |
| 55 | + } |
| 56 | + return false; |
| 57 | + }, |
| 58 | + }; |
| 59 | + |
| 60 | + if (!config.apiKey) { |
| 61 | + const { key } = await prompts( |
| 62 | + { |
| 63 | + type: "text", |
| 64 | + name: "key", |
| 65 | + message: askModels |
| 66 | + ? "Please provide your LLMHub API key (or leave blank to use T_SYSTEMS_LLMHUB_API_KEY env variable):" |
| 67 | + : "Please provide your LLMHub API key (leave blank to skip):", |
| 68 | + validate: (value: string) => { |
| 69 | + if (askModels && !value) { |
| 70 | + if (process.env.T_SYSTEMS_LLMHUB_API_KEY) { |
| 71 | + return true; |
| 72 | + } |
| 73 | + return "T_SYSTEMS_LLMHUB_API_KEY env variable is not set - key is required"; |
| 74 | + } |
| 75 | + return true; |
| 76 | + }, |
| 77 | + }, |
| 78 | + questionHandlers, |
| 79 | + ); |
| 80 | + config.apiKey = key || process.env.T_SYSTEMS_LLMHUB_API_KEY; |
| 81 | + } |
| 82 | + |
| 83 | + // use default model values in CI or if user should not be asked |
| 84 | + const useDefaults = ciInfo.isCI || !askModels; |
| 85 | + if (!useDefaults) { |
| 86 | + const { model } = await prompts( |
| 87 | + { |
| 88 | + type: "select", |
| 89 | + name: "model", |
| 90 | + message: "Which LLM model would you like to use?", |
| 91 | + choices: await getAvailableModelChoices(false, config.apiKey), |
| 92 | + initial: 0, |
| 93 | + }, |
| 94 | + questionHandlers, |
| 95 | + ); |
| 96 | + config.model = model; |
| 97 | + |
| 98 | + const { embeddingModel } = await prompts( |
| 99 | + { |
| 100 | + type: "select", |
| 101 | + name: "embeddingModel", |
| 102 | + message: "Which embedding model would you like to use?", |
| 103 | + choices: await getAvailableModelChoices(true, config.apiKey), |
| 104 | + initial: 0, |
| 105 | + }, |
| 106 | + questionHandlers, |
| 107 | + ); |
| 108 | + config.embeddingModel = embeddingModel; |
| 109 | + config.dimensions = getDimensions(embeddingModel); |
| 110 | + } |
| 111 | + |
| 112 | + return config; |
| 113 | +} |
| 114 | + |
| 115 | +async function getAvailableModelChoices( |
| 116 | + selectEmbedding: boolean, |
| 117 | + apiKey?: string, |
| 118 | +) { |
| 119 | + if (!apiKey) { |
| 120 | + throw new Error("Need LLMHub key to retrieve model choices"); |
| 121 | + } |
| 122 | + const isLLMModel = (modelId: string) => { |
| 123 | + return LLMHUB_MODELS.includes(modelId); |
| 124 | + }; |
| 125 | + |
| 126 | + const isEmbeddingModel = (modelId: string) => { |
| 127 | + return LLMHUB_EMBEDDING_MODELS.includes(modelId); |
| 128 | + }; |
| 129 | + |
| 130 | + const spinner = ora("Fetching available models").start(); |
| 131 | + try { |
| 132 | + const response = await got(`${TSYSTEMS_LLMHUB_API_URL}/models`, { |
| 133 | + headers: { |
| 134 | + Authorization: "Bearer " + apiKey, |
| 135 | + }, |
| 136 | + timeout: 5000, |
| 137 | + responseType: "json", |
| 138 | + }); |
| 139 | + const data: any = await response.body; |
| 140 | + spinner.stop(); |
| 141 | + return data.data |
| 142 | + .filter((model: any) => |
| 143 | + selectEmbedding ? isEmbeddingModel(model.id) : isLLMModel(model.id), |
| 144 | + ) |
| 145 | + .map((el: any) => { |
| 146 | + return { |
| 147 | + title: el.id, |
| 148 | + value: el.id, |
| 149 | + }; |
| 150 | + }); |
| 151 | + } catch (error) { |
| 152 | + spinner.stop(); |
| 153 | + if ((error as any).response?.statusCode === 401) { |
| 154 | + console.log( |
| 155 | + red( |
| 156 | + "Invalid LLMHub API key provided! Please provide a valid key and try again!", |
| 157 | + ), |
| 158 | + ); |
| 159 | + } else { |
| 160 | + console.log(red("Request failed: " + error)); |
| 161 | + } |
| 162 | + process.exit(1); |
| 163 | + } |
| 164 | +} |
| 165 | + |
| 166 | +function getDimensions(modelName: string) { |
| 167 | + // Assuming dimensions similar to OpenAI for simplicity. Update if different. |
| 168 | + return modelName === "text-embedding-004" ? 768 : 1536; |
| 169 | +} |
0 commit comments