-
Notifications
You must be signed in to change notification settings - Fork 2.1k
/
baiduwenxin.ts
709 lines (628 loc) Β· 19.1 KB
/
baiduwenxin.ts
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
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
import {
BaseChatModel,
type BaseChatModelParams,
} from "@langchain/core/language_models/chat_models";
import {
AIMessage,
AIMessageChunk,
BaseMessage,
ChatMessage,
} from "@langchain/core/messages";
import {
ChatGeneration,
ChatGenerationChunk,
ChatResult,
} from "@langchain/core/outputs";
import { CallbackManagerForLLMRun } from "@langchain/core/callbacks/manager";
import { getEnvironmentVariable } from "@langchain/core/utils/env";
import { convertEventStreamToIterableReadableDataStream } from "../utils/event_source_parse.js";
/**
* Type representing the role of a message in the Wenxin chat model.
*/
export type WenxinMessageRole = "assistant" | "user";
/**
* Interface representing a message in the Wenxin chat model.
*/
interface WenxinMessage {
role: WenxinMessageRole;
content: string;
}
/**
* Interface representing the usage of tokens in a chat completion.
*/
interface TokenUsage {
completionTokens?: number;
promptTokens?: number;
totalTokens?: number;
}
/**
* Interface representing a request for a chat completion.
*/
interface ChatCompletionRequest {
messages: WenxinMessage[];
stream?: boolean;
user_id?: string;
temperature?: number;
top_p?: number;
penalty_score?: number;
system?: string;
}
/**
* Interface representing a response from a chat completion.
*/
interface ChatCompletionResponse {
id: string;
object: string;
created: number;
result: string;
need_clear_history: boolean;
usage: TokenUsage;
}
/**
* Interface defining the input to the ChatBaiduWenxin class.
*/
declare interface BaiduWenxinChatInput {
/**
* Model name to use. Available options are: ERNIE-Bot, ERNIE-Bot-turbo, ERNIE-Bot-4
* Alias for `model`
* @default "ERNIE-Bot-turbo"
*/
modelName: string;
/** Model name to use. Available options are: ERNIE-Bot, ERNIE-Bot-turbo, ERNIE-Bot-4
* @default "ERNIE-Bot-turbo"
*/
model: string;
/** Whether to stream the results or not. Defaults to false. */
streaming?: boolean;
/** Messages to pass as a prefix to the prompt */
prefixMessages?: WenxinMessage[];
/**
* ID of the end-user who made requests.
*/
userId?: string;
/**
* API key to use when making requests. Defaults to the value of
* `BAIDU_API_KEY` environment variable.
* Alias for `apiKey`
*/
baiduApiKey?: string;
/**
* API key to use when making requests. Defaults to the value of
* `BAIDU_API_KEY` environment variable.
*/
apiKey?: string;
/**
* Secret key to use when making requests. Defaults to the value of
* `BAIDU_SECRET_KEY` environment variable.
*/
baiduSecretKey?: string;
/** Amount of randomness injected into the response. Ranges
* from 0 to 1 (0 is not included). Use temp closer to 0 for analytical /
* multiple choice, and temp closer to 1 for creative
* and generative tasks. Defaults to 0.95.
*/
temperature?: number;
/** Total probability mass of tokens to consider at each step. Range
* from 0 to 1.0. Defaults to 0.8.
*/
topP?: number;
/** Penalizes repeated tokens according to frequency. Range
* from 1.0 to 2.0. Defaults to 1.0.
*/
penaltyScore?: number;
}
/**
* Interface maps model names and their API endpoints.
*/
interface Models {
[key: string]: string;
}
/**
* Function that extracts the custom role of a generic chat message.
* @param message Chat message from which to extract the custom role.
* @returns The custom role of the chat message.
*/
function extractGenericMessageCustomRole(message: ChatMessage) {
if (message.role !== "assistant" && message.role !== "user") {
console.warn(`Unknown message role: ${message.role}`);
}
return message.role as WenxinMessageRole;
}
/**
* Function that converts a base message to a Wenxin message role.
* @param message Base message to convert.
* @returns The Wenxin message role.
*/
function messageToWenxinRole(message: BaseMessage): WenxinMessageRole {
const type = message._getType();
switch (type) {
case "ai":
return "assistant";
case "human":
return "user";
case "system":
throw new Error("System messages should not be here");
case "function":
throw new Error("Function messages not supported");
case "generic": {
if (!ChatMessage.isInstance(message))
throw new Error("Invalid generic chat message");
return extractGenericMessageCustomRole(message);
}
default:
throw new Error(`Unknown message type: ${type}`);
}
}
/**
* Wrapper around Baidu ERNIE large language models that use the Chat endpoint.
*
* To use you should have the `BAIDU_API_KEY` and `BAIDU_SECRET_KEY`
* environment variable set.
*
* @augments BaseLLM
* @augments BaiduERNIEInput
* @example
* ```typescript
* const ernieTurbo = new ChatBaiduWenxin({
* apiKey: "YOUR-API-KEY",
* baiduSecretKey: "YOUR-SECRET-KEY",
* });
*
* const ernie = new ChatBaiduWenxin({
* model: "ERNIE-Bot",
* temperature: 1,
* apiKey: "YOUR-API-KEY",
* baiduSecretKey: "YOUR-SECRET-KEY",
* });
*
* const messages = [new HumanMessage("Hello")];
*
* let res = await ernieTurbo.call(messages);
*
* res = await ernie.call(messages);
* ```
*/
export class ChatBaiduWenxin
extends BaseChatModel
implements BaiduWenxinChatInput
{
static lc_name() {
return "ChatBaiduWenxin";
}
get callKeys(): string[] {
return ["stop", "signal", "options"];
}
get lc_secrets(): { [key: string]: string } | undefined {
return {
baiduApiKey: "BAIDU_API_KEY",
apiKey: "BAIDU_API_KEY",
baiduSecretKey: "BAIDU_SECRET_KEY",
};
}
get lc_aliases(): { [key: string]: string } | undefined {
return undefined;
}
lc_serializable = true;
baiduApiKey?: string;
apiKey?: string;
baiduSecretKey?: string;
accessToken: string;
streaming = false;
prefixMessages?: WenxinMessage[];
userId?: string;
modelName = "ERNIE-Bot-turbo";
model = "ERNIE-Bot-turbo";
apiUrl: string;
temperature?: number | undefined;
topP?: number | undefined;
penaltyScore?: number | undefined;
constructor(fields?: Partial<BaiduWenxinChatInput> & BaseChatModelParams) {
super(fields ?? {});
this.baiduApiKey =
fields?.apiKey ??
fields?.baiduApiKey ??
getEnvironmentVariable("BAIDU_API_KEY");
if (!this.baiduApiKey) {
throw new Error("Baidu API key not found");
}
this.apiKey = this.baiduApiKey;
this.baiduSecretKey =
fields?.baiduSecretKey ?? getEnvironmentVariable("BAIDU_SECRET_KEY");
if (!this.baiduSecretKey) {
throw new Error("Baidu Secret key not found");
}
this.streaming = fields?.streaming ?? this.streaming;
this.prefixMessages = fields?.prefixMessages ?? this.prefixMessages;
this.userId = fields?.userId ?? this.userId;
this.temperature = fields?.temperature ?? this.temperature;
this.topP = fields?.topP ?? this.topP;
this.penaltyScore = fields?.penaltyScore ?? this.penaltyScore;
this.modelName = fields?.model ?? fields?.modelName ?? this.model;
this.model = this.modelName;
const models: Models = {
"ERNIE-Bot": "completions",
"ERNIE-Bot-turbo": "eb-instant",
"ERNIE-Bot-4": "completions_pro",
"ERNIE-Speed-8K": "ernie_speed",
"ERNIE-Speed-128K": "ernie-speed-128k",
"ERNIE-4.0-8K": "completions_pro",
"ERNIE-4.0-8K-Preview": "ernie-4.0-8k-preview",
"ERNIE-3.5-8K": "completions",
"ERNIE-3.5-8K-Preview": "ernie-3.5-8k-preview",
"ERNIE-Lite-8K": "eb-instant",
"ERNIE-Tiny-8K": "ernie-tiny-8k",
"ERNIE-Character-8K": "ernie-char-8k",
"ERNIE Speed-AppBuilder": "ai_apaas",
};
if (this.model in models) {
this.apiUrl = `https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/${
models[this.model]
}`;
} else {
throw new Error(`Invalid model name: ${this.model}`);
}
}
/**
* Method that retrieves the access token for making requests to the Baidu
* API.
* @param options Optional parsed call options.
* @returns The access token for making requests to the Baidu API.
*/
async getAccessToken(options?: this["ParsedCallOptions"]) {
const url = `https://aip.baidubce.com/oauth/2.0/token?grant_type=client_credentials&client_id=${this.apiKey}&client_secret=${this.baiduSecretKey}`;
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
Accept: "application/json",
},
signal: options?.signal,
});
if (!response.ok) {
const text = await response.text();
const error = new Error(
`Baidu get access token failed with status code ${response.status}, response: ${text}`
);
// eslint-disable-next-line @typescript-eslint/no-explicit-any
(error as any).response = response;
throw error;
}
const json = await response.json();
return json.access_token;
}
/**
* Get the parameters used to invoke the model
*/
invocationParams(): Omit<ChatCompletionRequest, "messages"> {
return {
stream: this.streaming,
user_id: this.userId,
temperature: this.temperature,
top_p: this.topP,
penalty_score: this.penaltyScore,
};
}
/**
* Get the identifying parameters for the model
*/
identifyingParams() {
return {
model_name: this.model,
...this.invocationParams(),
};
}
private _ensureMessages(messages: BaseMessage[]): WenxinMessage[] {
return messages.map((message) => ({
role: messageToWenxinRole(message),
content: message.text,
}));
}
/** @ignore */
async _generate(
messages: BaseMessage[],
options?: this["ParsedCallOptions"],
runManager?: CallbackManagerForLLMRun
): Promise<ChatResult> {
const tokenUsage: TokenUsage = {};
const params = this.invocationParams();
// Wenxin requires the system message to be put in the params, not messages array
const systemMessage = messages.find(
(message) => message._getType() === "system"
);
if (systemMessage) {
// eslint-disable-next-line no-param-reassign
messages = messages.filter((message) => message !== systemMessage);
params.system = systemMessage.text;
}
const messagesMapped = this._ensureMessages(messages);
const data = params.stream
? await new Promise<ChatCompletionResponse>((resolve, reject) => {
let response: ChatCompletionResponse;
let rejected = false;
let resolved = false;
this.completionWithRetry(
{
...params,
messages: messagesMapped,
},
true,
options?.signal,
(event) => {
const data = JSON.parse(event.data);
if (data?.error_code) {
if (rejected) {
return;
}
rejected = true;
reject(new Error(data?.error_msg));
return;
}
const message = data as {
id: string;
object: string;
created: number;
sentence_id?: number;
is_end: boolean;
result: string;
need_clear_history: boolean;
usage: TokenUsage;
};
// on the first message set the response properties
if (!response) {
response = {
id: message.id,
object: message.object,
created: message.created,
result: message.result,
need_clear_history: message.need_clear_history,
usage: message.usage,
};
} else {
response.result += message.result;
response.created = message.created;
response.need_clear_history = message.need_clear_history;
response.usage = message.usage;
}
// TODO this should pass part.index to the callback
// when that's supported there
// eslint-disable-next-line no-void
void runManager?.handleLLMNewToken(message.result ?? "");
if (message.is_end) {
if (resolved || rejected) {
return;
}
resolved = true;
resolve(response);
}
}
).catch((error) => {
if (!rejected) {
rejected = true;
reject(error);
}
});
})
: await this.completionWithRetry(
{
...params,
messages: messagesMapped,
},
false,
options?.signal
).then((data) => {
if (data?.error_code) {
throw new Error(data?.error_msg);
}
return data;
});
const {
completion_tokens: completionTokens,
prompt_tokens: promptTokens,
total_tokens: totalTokens,
} = data.usage ?? {};
if (completionTokens) {
tokenUsage.completionTokens =
(tokenUsage.completionTokens ?? 0) + completionTokens;
}
if (promptTokens) {
tokenUsage.promptTokens = (tokenUsage.promptTokens ?? 0) + promptTokens;
}
if (totalTokens) {
tokenUsage.totalTokens = (tokenUsage.totalTokens ?? 0) + totalTokens;
}
const generations: ChatGeneration[] = [];
const text = data.result ?? "";
generations.push({
text,
message: new AIMessage(text),
});
return {
generations,
llmOutput: { tokenUsage },
};
}
/** @ignore */
async completionWithRetry(
request: ChatCompletionRequest,
stream: boolean,
signal?: AbortSignal,
onmessage?: (event: MessageEvent) => void
) {
// The first run will get the accessToken
if (!this.accessToken) {
this.accessToken = await this.getAccessToken();
}
const findFirstNewlineIndex = (data: Uint8Array) => {
for (let i = 0; i < data.length; ) {
if (data[i] === 10) return i;
if ((data[i] & 0b11100000) === 0b11000000) {
i += 2;
} else if ((data[i] & 0b11110000) === 0b11100000) {
i += 3;
} else if ((data[i] & 0b11111000) === 0b11110000) {
i += 4;
} else {
i += 1;
}
}
return -1;
};
const makeCompletionRequest = async () => {
const url = `${this.apiUrl}?access_token=${this.accessToken}`;
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify(request),
signal,
});
if (!stream) {
return response.json();
} else {
if (response.body) {
// response will not be a stream if an error occurred
if (
!response.headers
.get("content-type")
?.startsWith("text/event-stream")
) {
onmessage?.(
new MessageEvent("message", {
data: await response.text(),
})
);
return;
}
const reader = response.body.getReader();
const decoder = new TextDecoder("utf-8");
let dataArrayBuffer = new Uint8Array(0);
let continueReading = true;
while (continueReading) {
const { done, value } = await reader.read();
if (done) {
continueReading = false;
break;
}
// merge the data first then decode in case of the Chinese characters are split between chunks
const mergedArray = new Uint8Array(
dataArrayBuffer.length + value.length
);
mergedArray.set(dataArrayBuffer);
mergedArray.set(value, dataArrayBuffer.length);
dataArrayBuffer = mergedArray;
let continueProcessing = true;
while (continueProcessing) {
const newlineIndex = findFirstNewlineIndex(dataArrayBuffer);
if (newlineIndex === -1) {
continueProcessing = false;
break;
}
const lineArrayBuffer = dataArrayBuffer.slice(
0,
findFirstNewlineIndex(dataArrayBuffer)
);
const line = decoder.decode(lineArrayBuffer);
dataArrayBuffer = dataArrayBuffer.slice(
findFirstNewlineIndex(dataArrayBuffer) + 1
);
if (line.startsWith("data:")) {
const event = new MessageEvent("message", {
data: line.slice("data:".length).trim(),
});
onmessage?.(event);
}
}
}
}
}
};
return this.caller.call(makeCompletionRequest);
}
private async getFullApiUrl() {
if (!this.accessToken) {
this.accessToken = await this.getAccessToken();
}
return `${this.apiUrl}?access_token=${this.accessToken}`;
}
private async createWenxinStream(
request: ChatCompletionRequest,
signal?: AbortSignal
) {
const url = await this.getFullApiUrl();
const response = await fetch(url, {
method: "POST",
headers: {
Accept: "text/event-stream",
"Content-Type": "application/json",
},
body: JSON.stringify(request),
signal,
});
if (!response.body) {
throw new Error(
"Could not begin Wenxin stream. Please check the given URL and try again."
);
}
return convertEventStreamToIterableReadableDataStream(response.body);
}
private _deserialize(json: string) {
try {
return JSON.parse(json);
} catch (e) {
console.warn(`Received a non-JSON parseable chunk: ${json}`);
}
}
async *_streamResponseChunks(
messages: BaseMessage[],
options?: this["ParsedCallOptions"],
runManager?: CallbackManagerForLLMRun
): AsyncGenerator<ChatGenerationChunk> {
const parameters = {
...this.invocationParams(),
stream: true,
};
// Wenxin requires the system message to be put in the params, not messages array
const systemMessage = messages.find(
(message) => message._getType() === "system"
);
if (systemMessage) {
// eslint-disable-next-line no-param-reassign
messages = messages.filter((message) => message !== systemMessage);
parameters.system = systemMessage.text;
}
const messagesMapped = this._ensureMessages(messages);
const stream = await this.caller.call(async () =>
this.createWenxinStream(
{
...parameters,
messages: messagesMapped,
},
options?.signal
)
);
for await (const chunk of stream) {
const deserializedChunk = this._deserialize(chunk);
const { result, is_end, id } = deserializedChunk;
yield new ChatGenerationChunk({
text: result,
message: new AIMessageChunk({ content: result }),
generationInfo: is_end
? {
is_end,
request_id: id,
usage: chunk.usage,
}
: undefined,
});
await runManager?.handleLLMNewToken(result);
}
}
_llmType() {
return "baiduwenxin";
}
/** @ignore */
_combineLLMOutput() {
return [];
}
}