Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 | 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x | /**
* @license
* Copyright 2025 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
// DISCLAIMER: This is a copied version of https://github.com/googleapis/js-genai/blob/main/src/chats.ts with the intention of working around a key bug
// where function responses are not treated as "valid" responses: https://b.corp.google.com/issues/420354090
import {
GenerateContentResponse,
type Content,
type GenerateContentConfig,
type SendMessageParameters,
type Part,
ApiError,
} from '@google/genai';
import { createUserContent } from '@google/genai';
import { retryWithBackoff } from '../utils/retry';
import type { Config } from '../config/config';
import { getEffectiveModel } from '../config/models';
export enum StreamEventType {
/** A regular content chunk from the API. */
CHUNK = 'chunk',
/** A signal that a retry is about to happen. The UI should discard any partial
* content from the attempt that just failed. */
RETRY = 'retry',
}
export type StreamEvent =
| { type: StreamEventType.CHUNK; value: GenerateContentResponse }
| { type: StreamEventType.RETRY };
/**
* Options for retrying due to invalid content from the model.
*/
interface ContentRetryOptions {
/** Total number of attempts to make (1 initial + N retries). */
maxAttempts: number;
/** The base delay in milliseconds for linear backoff. */
initialDelayMs: number;
}
const INVALID_CONTENT_RETRY_OPTIONS: ContentRetryOptions = {
maxAttempts: 2, // 1 initial call + 1 retry
initialDelayMs: 500,
};
/**
* Returns true if the response is valid, false otherwise.
*/
function isValidResponse(response: GenerateContentResponse): boolean {
Iif (response.candidates === undefined || response.candidates.length === 0) {
return false;
}
const content = response.candidates[0]?.content;
Iif (content === undefined) {
return false;
}
return isValidContent(content);
}
export function isValidNonThoughtTextPart(part: Part): boolean {
return (
typeof part.text === 'string' &&
!part.thought &&
// Technically, the model should never generate parts that have text and
// any of these but we don't trust them so check anyways.
!part.inlineData &&
!part.fileData
);
}
function isValidContent(content: Content): boolean {
Iif (content.parts === undefined || content.parts.length === 0) {
return false;
}
for (const part of content.parts) {
Iif (part === undefined || Object.keys(part).length === 0) {
return false;
}
Iif (!part.thought && part.text !== undefined && part.text === '') {
return false;
}
}
return true;
}
/**
* Validates the history contains the correct roles.
*
* @throws Error if the history does not start with a user turn.
* @throws Error if the history contains an invalid role.
*/
function validateHistory(history: Content[]) {
for (const content of history) {
Iif (content.role !== 'user' && content.role !== 'model') {
throw new Error(`Role must be user or model, but got ${content.role}.`);
}
}
}
/**
* Extracts the curated (valid) history from a comprehensive history.
*
* @remarks
* The model may sometimes generate invalid or empty contents(e.g., due to safety
* filters or recitation). Extracting valid turns from the history
* ensures that subsequent requests could be accepted by the model.
*/
function extractCuratedHistory(comprehensiveHistory: Content[]): Content[] {
Iif (comprehensiveHistory === undefined || comprehensiveHistory.length === 0) {
return [];
}
const curatedHistory: Content[] = [];
const length = comprehensiveHistory.length;
let i = 0;
while (i < length) {
if (comprehensiveHistory[i].role === 'user') {
curatedHistory.push(comprehensiveHistory[i]);
i++;
} else {
const modelOutput: Content[] = [];
let isValid = true;
while (i < length && comprehensiveHistory[i].role === 'model') {
modelOutput.push(comprehensiveHistory[i]);
Iif (isValid && !isValidContent(comprehensiveHistory[i])) {
isValid = false;
}
i++;
}
Iif (isValid) {
curatedHistory.push(...modelOutput);
}
}
}
return curatedHistory;
}
/**
* Custom error to signal that a stream completed with invalid content,
* which should trigger a retry.
*/
export class InvalidStreamError extends Error {
readonly type: 'NO_FINISH_REASON' | 'NO_RESPONSE_TEXT';
constructor(message: string, type: 'NO_FINISH_REASON' | 'NO_RESPONSE_TEXT') {
super(message);
this.name = 'InvalidStreamError';
this.type = type;
}
}
/**
* Chat session that enables sending messages to the model with previous
* conversation context.
*
* @remarks
* The session maintains all the turns between user and model.
*/
export class GeminiChat {
// A promise to represent the current state of the message being sent to the
// model.
private sendPromise: Promise<void> = Promise.resolve();
constructor(
private readonly config: Config,
private readonly generationConfig: GenerateContentConfig = {},
private history: Content[] = [],
) {
validateHistory(history);
}
setSystemInstruction(sysInstr: string) {
this.generationConfig.systemInstruction = sysInstr;
}
async sendMessageStream(
model: string,
params: SendMessageParameters,
prompt_id: string,
): Promise<AsyncGenerator<StreamEvent>> {
await this.sendPromise;
let streamDoneResolver: () => void;
const streamDonePromise = new Promise<void>((resolve) => {
streamDoneResolver = resolve;
});
this.sendPromise = streamDonePromise;
const userContent = createUserContent(params.message);
// Add user content to history ONCE before any attempts.
this.history.push(userContent);
const requestContents = this.getHistory(true);
// eslint-disable-next-line @typescript-eslint/no-this-alias
const self = this;
return (async function* () {
try {
let lastError: unknown = new Error('Request failed after all retries.');
for (
let attempt = 0;
attempt < INVALID_CONTENT_RETRY_OPTIONS.maxAttempts;
attempt++
) {
try {
Iif (attempt > 0) {
yield { type: StreamEventType.RETRY };
}
const stream = await self.makeApiCallAndProcessStream(
model,
requestContents,
params,
prompt_id,
);
for await (const chunk of stream) {
yield { type: StreamEventType.CHUNK, value: chunk };
}
lastError = null;
break;
} catch (error) {
lastError = error;
const isContentError = error instanceof InvalidStreamError;
Iif (isContentError) {
// Check if we have more attempts left.
Iif (attempt < INVALID_CONTENT_RETRY_OPTIONS.maxAttempts - 1) {
await new Promise((res) =>
setTimeout(
res,
INVALID_CONTENT_RETRY_OPTIONS.initialDelayMs *
(attempt + 1),
),
);
continue;
}
}
break;
}
}
Iif (lastError) {
Iif (self.history[self.history.length - 1] === userContent) {
self.history.pop();
}
throw lastError;
}
} finally {
streamDoneResolver!();
}
})();
}
private async makeApiCallAndProcessStream(
model: string,
requestContents: Content[],
params: SendMessageParameters,
prompt_id: string,
): Promise<AsyncGenerator<GenerateContentResponse>> {
const apiCall = () => {
const modelToUse = getEffectiveModel(model);
return this.config.getContentGenerator().generateContentStream(
{
model: modelToUse,
contents: requestContents,
config: { ...this.generationConfig, ...params.config },
},
prompt_id,
);
};
const streamResponse = await retryWithBackoff(apiCall, {
shouldRetryOnError: (error: unknown) => {
Iif (error instanceof ApiError && error.message) {
Iif (error.status === 400) return false;
Iif (isSchemaDepthError(error.message)) return false;
Iif (error.status === 429) return true;
Iif (error.status >= 500 && error.status < 600) return true;
}
return false;
},
authType: this.config.getContentGeneratorConfig()?.authType,
});
return this.processStreamResponse(model, streamResponse);
}
getHistory(curated: boolean = false): Content[] {
const history = curated
? extractCuratedHistory(this.history)
: this.history;
// Deep copy the history to avoid mutating the history outside of the
// chat session.
return structuredClone(history);
}
clearHistory(): void {
this.history = [];
}
addHistory(content: Content): void {
this.history.push(content);
}
setHistory(history: Content[]): void {
this.history = history;
}
private async *processStreamResponse(
model: string,
streamResponse: AsyncGenerator<GenerateContentResponse>,
): AsyncGenerator<GenerateContentResponse> {
const modelResponseParts: Part[] = [];
let hasFinishReason = false;
for await (const chunk of streamResponse) {
hasFinishReason =
chunk?.candidates?.some((candidate) => candidate.finishReason) ?? false;
Iif (isValidResponse(chunk)) {
const content = chunk.candidates?.[0]?.content;
Iif (content?.parts) {
modelResponseParts.push(
...content.parts.filter((part) => !part.thought),
);
}
}
yield chunk; // Yield every chunk to the UI immediately.
}
// String thoughts and consolidate text parts.
const consolidatedParts: Part[] = [];
for (const part of modelResponseParts) {
const lastPart = consolidatedParts[consolidatedParts.length - 1];
if (
lastPart?.text &&
isValidNonThoughtTextPart(lastPart) &&
isValidNonThoughtTextPart(part)
) {
lastPart.text += part.text;
} else {
consolidatedParts.push(part);
}
}
const responseText = consolidatedParts
.filter((part) => part.text)
.map((part) => part.text)
.join('')
.trim();
Iif (!hasFinishReason || !responseText) {
if (!hasFinishReason) {
throw new InvalidStreamError(
'Model stream ended without a finish reason.',
'NO_FINISH_REASON',
);
} else {
throw new InvalidStreamError(
'Model stream ended with empty response text.',
'NO_RESPONSE_TEXT',
);
}
}
this.history.push({ role: 'model', parts: consolidatedParts });
}
}
/** Visible for Testing */
export function isSchemaDepthError(errorMessage: string): boolean {
return errorMessage.includes('maximum schema depth exceeded');
}
export function isInvalidArgumentError(errorMessage: string): boolean {
return errorMessage.includes('Request contains an invalid argument');
} |