|
| 1 | +/** |
| 2 | + * @license |
| 3 | + * Copyright 2023 Google LLC |
| 4 | + * |
| 5 | + * Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | + * you may not use this file except in compliance with the License. |
| 7 | + * You may obtain a copy of the License at |
| 8 | + * |
| 9 | + * https://www.apache.org/licenses/LICENSE-2.0 |
| 10 | + * |
| 11 | + * Unless required by applicable law or agreed to in writing, software |
| 12 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | + * See the License for the specific language governing permissions and |
| 15 | + * limitations under the License. |
| 16 | + */ |
| 17 | + |
| 18 | +/* tslint:disable */ |
| 19 | +import {GoogleAuth} from 'google-auth-library'; |
| 20 | + |
| 21 | +import {emptyGenerator, processStream} from './process_stream'; |
| 22 | +import {Content, GenerateContentParams, GenerateContentRequest, GenerateContentResult, GenerationConfig, Part, SafetySetting} from './types/content'; |
| 23 | +import {postRequest} from './util'; |
| 24 | + |
| 25 | +// TODO: update this when model names are available |
| 26 | +// const SUPPORTED_MODELS: Array<string> = ['text-bison@001']; |
| 27 | + |
| 28 | +/** |
| 29 | + * Base class for authenticating to Vertex, creates the preview namespace. |
| 30 | + */ |
| 31 | +export class VertexAI { |
| 32 | + public preview: VertexAI_Internal; |
| 33 | + |
| 34 | + constructor( |
| 35 | + project: string, |
| 36 | + location: string, |
| 37 | + apiKey: string, // TODO: remove when we switch to Vertex endpoint |
| 38 | + apiEndpoint?: string, |
| 39 | + ) { |
| 40 | + this.preview = |
| 41 | + new VertexAI_Internal(project, location, apiKey, apiEndpoint); |
| 42 | + } |
| 43 | +} |
| 44 | + |
| 45 | +/** |
| 46 | + * VertexAI class implementation |
| 47 | + */ |
| 48 | +export class VertexAI_Internal { |
| 49 | + protected googleAuth: GoogleAuth = new GoogleAuth( |
| 50 | + {scopes: 'https://www.googleapis.com/auth/cloud-platform'}); |
| 51 | + private tokenInternal?: string; |
| 52 | + |
| 53 | + /** |
| 54 | + * API client for authenticating to Vertex |
| 55 | + * @param project The Google Cloud project to use for the request |
| 56 | + * @param location The Google Cloud project location to use for the |
| 57 | + * request |
| 58 | + * @param apiEndpoint The base Vertex AI endpoint to use for the request. If |
| 59 | + * not provided, the default regionalized endpoint (i.e. |
| 60 | + * us-central1-aiplatform.googleapis.com) will be used. |
| 61 | + */ |
| 62 | + constructor( |
| 63 | + protected readonly project: string, |
| 64 | + protected readonly location: string, |
| 65 | + protected readonly apiKey: |
| 66 | + string, // TODO: remove when we switch to Vertex endpoint |
| 67 | + protected readonly apiEndpoint?: string, |
| 68 | + ) {} |
| 69 | + |
| 70 | + /** |
| 71 | + * Gets an authentication token for making Vertex REST API requests. |
| 72 | + * @param vertex The VertexAI instance. |
| 73 | + */ |
| 74 | + // TODO: change the `any` type below to be more specific |
| 75 | + protected get token(): Promise<any>|string { |
| 76 | + if (this.tokenInternal) { |
| 77 | + return this.tokenInternal; |
| 78 | + } |
| 79 | + // Generate a new token if it hasn't been set |
| 80 | + // TODO: add error handling here |
| 81 | + const token = Promise.resolve(this.googleAuth.getAccessToken()); |
| 82 | + return token; |
| 83 | + } |
| 84 | + |
| 85 | + /** |
| 86 | + * Make a generateContent request. |
| 87 | + * @param request A GenerateContentRequest object with the request contents. |
| 88 | + * @return The GenerateContentResponse object with the response candidates. |
| 89 | + * |
| 90 | + * NOTE: this method is stubbed in postRequest for now until the service is |
| 91 | + * available. |
| 92 | + */ |
| 93 | + async generateContent(request: GenerateContentParams): |
| 94 | + Promise<GenerateContentResult> { |
| 95 | + const publisherModelEndpoint = `publishers/google/models/${request.model}`; |
| 96 | + |
| 97 | + const generateContentRequest: GenerateContentRequest = { |
| 98 | + model: request.model, |
| 99 | + contents: request.contents, |
| 100 | + generation_config: request.generation_config, |
| 101 | + safety_settings: request.safety_settings, |
| 102 | + } |
| 103 | + |
| 104 | + let response; |
| 105 | + try { |
| 106 | + response = await postRequest({ |
| 107 | + region: this.location, |
| 108 | + project: this.project, |
| 109 | + resourcePath: publisherModelEndpoint, |
| 110 | + resourceMethod: request.stream ? 'streamGenerateContent' : |
| 111 | + 'generateContent', |
| 112 | + token: await this.token, |
| 113 | + data: generateContentRequest, |
| 114 | + apiKey: this.apiKey, |
| 115 | + apiEndpoint: this.apiEndpoint, |
| 116 | + }); |
| 117 | + if (response === undefined) { |
| 118 | + throw new Error('did not get a valid response.') |
| 119 | + } |
| 120 | + if (!response.ok) { |
| 121 | + throw new Error(`${response.status} ${response.statusText}`) |
| 122 | + } |
| 123 | + } catch (e) { |
| 124 | + console.log(e); |
| 125 | + } |
| 126 | + |
| 127 | + const streamResult = processStream(response); |
| 128 | + |
| 129 | + if (request.stream === false && streamResult.stream !== undefined) { |
| 130 | + const responses = []; |
| 131 | + for await (const resp of streamResult.stream) { |
| 132 | + responses.push(resp); |
| 133 | + } |
| 134 | + return { |
| 135 | + stream: emptyGenerator(), |
| 136 | + responses, |
| 137 | + }; |
| 138 | + } else { |
| 139 | + // True or undefined (default true) |
| 140 | + return streamResult; |
| 141 | + } |
| 142 | + |
| 143 | + // TODO: handle streaming and non-streaming response here |
| 144 | + } |
| 145 | + |
| 146 | + startChat(request: StartChatParams): ChatSession { |
| 147 | + const startChatRequest = { |
| 148 | + model: request.model, |
| 149 | + project: this.project, |
| 150 | + location: this.location, |
| 151 | + history: request.history, |
| 152 | + generation_config: request.generation_config, |
| 153 | + safety_settings: request.safety_settings, |
| 154 | + _vertex_instance: this, |
| 155 | + }; |
| 156 | + |
| 157 | + return new ChatSession(startChatRequest); |
| 158 | + } |
| 159 | +} |
| 160 | + |
| 161 | +/** |
| 162 | + * Params to initiate a multiturn chat with the model via startChat |
| 163 | + */ |
| 164 | +export declare interface StartChatParams { |
| 165 | + model: string; |
| 166 | + history?: Content[]; |
| 167 | + safety_settings?: SafetySetting[]; |
| 168 | + generation_config?: GenerationConfig; |
| 169 | + stream?: boolean; |
| 170 | +} |
| 171 | + |
| 172 | +// StartChatSessionRequest and ChatSession are defined here instead of in |
| 173 | +// src/types to avoid a circular dependency issue due the dep on |
| 174 | +// VertexAI_Internal |
| 175 | + |
| 176 | +/** |
| 177 | + * All params passed to initiate multiturn chat via startChat |
| 178 | + */ |
| 179 | +export declare interface StartChatSessionRequest extends StartChatParams { |
| 180 | + project: string; |
| 181 | + location: string; |
| 182 | + _vertex_instance: VertexAI_Internal; |
| 183 | +} |
| 184 | + |
| 185 | +/** |
| 186 | + * Session for a multiturn chat with the model |
| 187 | + */ |
| 188 | +export class ChatSession { |
| 189 | + // Substitute apiKey for these in Labs |
| 190 | + private project: string; |
| 191 | + private location: string; |
| 192 | + |
| 193 | + private _history: Content[]; |
| 194 | + private _vertex_instance: VertexAI_Internal; |
| 195 | + |
| 196 | + |
| 197 | + model: string; |
| 198 | + generation_config?: GenerationConfig; |
| 199 | + safety_settings?: SafetySetting[]; |
| 200 | + |
| 201 | + get history(): Content[] { |
| 202 | + return this._history; |
| 203 | + } |
| 204 | + |
| 205 | + constructor(request: StartChatSessionRequest) { |
| 206 | + this.project = request.project; |
| 207 | + this.location = request.location; |
| 208 | + this.model = request.model; |
| 209 | + this._history = request.history ?? []; |
| 210 | + this._vertex_instance = request._vertex_instance; |
| 211 | + } |
| 212 | + |
| 213 | + async sendMessage(request: string| |
| 214 | + Array<string|Part>): Promise<GenerateContentResult> { |
| 215 | + // TODO: this is stubbed until the service is available |
| 216 | + let generateContentrequest: GenerateContentParams = { |
| 217 | + model: this.model, |
| 218 | + contents: [], |
| 219 | + safety_settings: this.safety_settings, |
| 220 | + generation_config: this.generation_config, |
| 221 | + }; |
| 222 | + |
| 223 | + let currentContent = []; |
| 224 | + |
| 225 | + if (typeof request === 'string') { |
| 226 | + currentContent = [{role: 'user', parts: [{text: request}]}]; |
| 227 | + } else if (Array.isArray(request)) { |
| 228 | + for (const item of request) { |
| 229 | + if (typeof item === 'string') { |
| 230 | + currentContent.push({role: 'user', parts: [{text: item}]}); |
| 231 | + } else { |
| 232 | + currentContent.push({role: 'user', parts: [item]}); |
| 233 | + } |
| 234 | + } |
| 235 | + }; |
| 236 | + |
| 237 | + generateContentrequest.contents = currentContent; |
| 238 | + const generateContentResponse = |
| 239 | + await this._vertex_instance.generateContent(generateContentrequest); |
| 240 | + // TODO: add error handling |
| 241 | + |
| 242 | + // First add the messages sent by the user |
| 243 | + for (const content of currentContent) { |
| 244 | + this._history.push(content); |
| 245 | + }; |
| 246 | + |
| 247 | + for (const result of generateContentResponse.responses) { |
| 248 | + for (const candidate of result.candidates) { |
| 249 | + this._history.push(candidate.content); |
| 250 | + } |
| 251 | + } |
| 252 | + return generateContentResponse; |
| 253 | + } |
| 254 | +} |
| 255 | + |
| 256 | +export { |
| 257 | + Content, |
| 258 | + GenerateContentParams, |
| 259 | + GenerateContentRequest, |
| 260 | + GenerateContentResult, |
| 261 | + GenerationConfig, |
| 262 | + Part, |
| 263 | + SafetySetting |
| 264 | +}; |
0 commit comments