-
Notifications
You must be signed in to change notification settings - Fork 2.1k
/
index.ts
308 lines (270 loc) Β· 8.01 KB
/
index.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
import { type Tiktoken } from "js-tiktoken/lite";
import type { OpenAI as OpenAIClient } from "openai";
import {
BaseCache,
BaseMessage,
BaseMessageLike,
BasePromptValue,
LLMResult,
MessageContent,
coerceMessageLikeToMessage,
} from "../schema/index.js";
import {
BaseCallbackConfig,
CallbackManager,
Callbacks,
} from "../callbacks/manager.js";
import { AsyncCaller, AsyncCallerParams } from "../util/async_caller.js";
import { getModelNameForTiktoken } from "./count_tokens.js";
import { encodingForModel } from "../util/tiktoken.js";
import { Runnable } from "../schema/runnable/index.js";
import { RunnableConfig } from "../schema/runnable/config.js";
import { StringPromptValue } from "../prompts/base.js";
import { ChatPromptValue } from "../prompts/chat.js";
import { InMemoryCache } from "../cache/index.js";
const getVerbosity = () => false;
export type SerializedLLM = {
_model: string;
_type: string;
// eslint-disable-next-line @typescript-eslint/no-explicit-any
} & Record<string, any>;
export interface BaseLangChainParams {
verbose?: boolean;
callbacks?: Callbacks;
tags?: string[];
metadata?: Record<string, unknown>;
}
/**
* Base class for language models, chains, tools.
*/
export abstract class BaseLangChain<
RunInput,
RunOutput,
CallOptions extends RunnableConfig = RunnableConfig
>
extends Runnable<RunInput, RunOutput, CallOptions>
implements BaseLangChainParams
{
/**
* Whether to print out response text.
*/
verbose: boolean;
callbacks?: Callbacks;
tags?: string[];
metadata?: Record<string, unknown>;
get lc_attributes(): { [key: string]: undefined } | undefined {
return {
callbacks: undefined,
verbose: undefined,
};
}
constructor(params: BaseLangChainParams) {
super(params);
this.verbose = params.verbose ?? getVerbosity();
this.callbacks = params.callbacks;
this.tags = params.tags ?? [];
this.metadata = params.metadata ?? {};
}
}
/**
* Base interface for language model parameters.
* A subclass of {@link BaseLanguageModel} should have a constructor that
* takes in a parameter that extends this interface.
*/
export interface BaseLanguageModelParams
extends AsyncCallerParams,
BaseLangChainParams {
/**
* @deprecated Use `callbacks` instead
*/
callbackManager?: CallbackManager;
cache?: BaseCache | boolean;
}
export interface BaseLanguageModelCallOptions extends BaseCallbackConfig {
/**
* Stop tokens to use for this call.
* If not provided, the default stop tokens for the model will be used.
*/
stop?: string[];
/**
* Timeout for this call in milliseconds.
*/
timeout?: number;
/**
* Abort signal for this call.
* If provided, the call will be aborted when the signal is aborted.
* @see https://developer.mozilla.org/en-US/docs/Web/API/AbortSignal
*/
signal?: AbortSignal;
}
export interface BaseFunctionCallOptions extends BaseLanguageModelCallOptions {
function_call?: OpenAIClient.Chat.ChatCompletionFunctionCallOption;
functions?: OpenAIClient.Chat.ChatCompletionCreateParams.Function[];
}
export type BaseLanguageModelInput =
| BasePromptValue
| string
| BaseMessageLike[];
/**
* Base class for language models.
*/
export abstract class BaseLanguageModel<
// eslint-disable-next-line @typescript-eslint/no-explicit-any
RunOutput = any,
CallOptions extends BaseLanguageModelCallOptions = BaseLanguageModelCallOptions
>
extends BaseLangChain<BaseLanguageModelInput, RunOutput, CallOptions>
implements BaseLanguageModelParams
{
declare CallOptions: CallOptions;
/**
* Keys that the language model accepts as call options.
*/
get callKeys(): string[] {
return ["stop", "timeout", "signal", "tags", "metadata", "callbacks"];
}
/**
* The async caller should be used by subclasses to make any async calls,
* which will thus benefit from the concurrency and retry logic.
*/
caller: AsyncCaller;
cache?: BaseCache;
constructor({
callbacks,
callbackManager,
...params
}: BaseLanguageModelParams) {
super({
callbacks: callbacks ?? callbackManager,
...params,
});
if (typeof params.cache === "object") {
this.cache = params.cache;
} else if (params.cache) {
this.cache = InMemoryCache.global();
} else {
this.cache = undefined;
}
this.caller = new AsyncCaller(params ?? {});
}
abstract generatePrompt(
promptValues: BasePromptValue[],
options?: string[] | CallOptions,
callbacks?: Callbacks
): Promise<LLMResult>;
abstract predict(
text: string,
options?: string[] | CallOptions,
callbacks?: Callbacks
): Promise<string>;
abstract predictMessages(
messages: BaseMessage[],
options?: string[] | CallOptions,
callbacks?: Callbacks
): Promise<BaseMessage>;
abstract _modelType(): string;
abstract _llmType(): string;
private _encoding?: Tiktoken;
async getNumTokens(content: MessageContent) {
// TODO: Figure out correct value.
if (typeof content !== "string") {
return 0;
}
// fallback to approximate calculation if tiktoken is not available
let numTokens = Math.ceil(content.length / 4);
if (!this._encoding) {
try {
this._encoding = await encodingForModel(
"modelName" in this
? getModelNameForTiktoken(this.modelName as string)
: "gpt2"
);
} catch (error) {
console.warn(
"Failed to calculate number of tokens, falling back to approximate count",
error
);
}
}
if (this._encoding) {
numTokens = this._encoding.encode(content).length;
}
return numTokens;
}
protected static _convertInputToPromptValue(
input: BaseLanguageModelInput
): BasePromptValue {
if (typeof input === "string") {
return new StringPromptValue(input);
} else if (Array.isArray(input)) {
return new ChatPromptValue(input.map(coerceMessageLikeToMessage));
} else {
return input;
}
}
/**
* Get the identifying parameters of the LLM.
*/
// eslint-disable-next-line @typescript-eslint/no-explicit-any
_identifyingParams(): Record<string, any> {
return {};
}
/**
* Create a unique cache key for a specific call to a specific language model.
* @param callOptions Call options for the model
* @returns A unique cache key.
*/
protected _getSerializedCacheKeyParametersForCall(
callOptions: CallOptions
): string {
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const params: Record<string, any> = {
...this._identifyingParams(),
...callOptions,
_type: this._llmType(),
_model: this._modelType(),
};
const filteredEntries = Object.entries(params).filter(
([_, value]) => value !== undefined
);
const serializedEntries = filteredEntries
.map(([key, value]) => `${key}:${JSON.stringify(value)}`)
.sort()
.join(",");
return serializedEntries;
}
/**
* @deprecated
* Return a json-like object representing this LLM.
*/
serialize(): SerializedLLM {
return {
...this._identifyingParams(),
_type: this._llmType(),
_model: this._modelType(),
};
}
/**
* @deprecated
* Load an LLM from a json-like object describing it.
*/
static async deserialize(data: SerializedLLM): Promise<BaseLanguageModel> {
const { _type, _model, ...rest } = data;
if (_model && _model !== "base_chat_model") {
throw new Error(`Cannot load LLM with model ${_model}`);
}
const Cls = {
openai: (await import("../chat_models/openai.js")).ChatOpenAI,
}[_type];
if (Cls === undefined) {
throw new Error(`Cannot load LLM with type ${_type}`);
}
return new Cls(rest);
}
}
/*
* Export utility functions for token calculations:
* - calculateMaxTokens: Calculate max tokens for a given model and prompt (the model context size - tokens in prompt).
* - getModelContextSize: Get the context size for a specific model.
*/
export { calculateMaxTokens, getModelContextSize } from "./count_tokens.js";