modelfusion 0.82.0 → 0.84.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +13 -10
- package/core/schema/UncheckedSchema.cjs +5 -1
- package/core/schema/UncheckedSchema.d.ts +1 -0
- package/core/schema/UncheckedSchema.js +3 -0
- package/model-function/AbstractModel.d.ts +1 -1
- package/model-function/generate-text/prompt-format/InstructionPrompt.d.ts +1 -1
- package/model-provider/automatic1111/Automatic1111ImageGenerationModel.cjs +1 -1
- package/model-provider/automatic1111/Automatic1111ImageGenerationModel.d.ts +1 -1
- package/model-provider/automatic1111/Automatic1111ImageGenerationModel.js +1 -1
- package/model-provider/index.cjs +1 -0
- package/model-provider/index.d.ts +1 -0
- package/model-provider/index.js +1 -0
- package/model-provider/llamacpp/LlamaCppBakLLaVA1Format.d.ts +2 -2
- package/model-provider/openai/AzureOpenAIApiConfiguration.cjs +1 -1
- package/model-provider/openai/AzureOpenAIApiConfiguration.d.ts +1 -1
- package/model-provider/openai/AzureOpenAIApiConfiguration.js +1 -1
- package/model-provider/openai/chat/AbstractOpenAIChatModel.cjs +228 -0
- package/model-provider/openai/chat/AbstractOpenAIChatModel.d.ts +467 -0
- package/model-provider/openai/chat/AbstractOpenAIChatModel.js +224 -0
- package/model-provider/openai/chat/OpenAIChatFunctionCallStructureGenerationModel.cjs +3 -3
- package/model-provider/openai/chat/OpenAIChatFunctionCallStructureGenerationModel.d.ts +6 -6
- package/model-provider/openai/chat/OpenAIChatFunctionCallStructureGenerationModel.js +1 -1
- package/model-provider/openai/chat/OpenAIChatModel.cjs +5 -218
- package/model-provider/openai/chat/OpenAIChatModel.d.ts +11 -460
- package/model-provider/openai/chat/OpenAIChatModel.js +4 -217
- package/model-provider/openai/chat/OpenAIChatPromptFormat.d.ts +2 -2
- package/model-provider/openai/index.cjs +1 -0
- package/model-provider/openai/index.d.ts +1 -0
- package/model-provider/openai/index.js +1 -0
- package/model-provider/openai-compatible/FireworksAIApiConfiguration.cjs +29 -0
- package/model-provider/openai-compatible/FireworksAIApiConfiguration.d.ts +18 -0
- package/model-provider/openai-compatible/FireworksAIApiConfiguration.js +25 -0
- package/model-provider/openai-compatible/OpenAICompatibleChatModel.cjs +100 -0
- package/model-provider/openai-compatible/OpenAICompatibleChatModel.d.ts +45 -0
- package/model-provider/openai-compatible/OpenAICompatibleChatModel.js +96 -0
- package/model-provider/openai-compatible/OpenAICompatibleFacade.cjs +30 -0
- package/model-provider/openai-compatible/OpenAICompatibleFacade.d.ts +24 -0
- package/model-provider/openai-compatible/OpenAICompatibleFacade.js +26 -0
- package/model-provider/openai-compatible/index.cjs +32 -0
- package/model-provider/openai-compatible/index.d.ts +3 -0
- package/model-provider/openai-compatible/index.js +3 -0
- package/model-provider/stability/StabilityImageGenerationModel.cjs +1 -1
- package/model-provider/stability/StabilityImageGenerationModel.d.ts +1 -1
- package/model-provider/stability/StabilityImageGenerationModel.js +1 -1
- package/package.json +1 -1
@@ -0,0 +1,467 @@
|
|
1
|
+
import { z } from "zod";
|
2
|
+
import { FunctionOptions } from "../../../core/FunctionOptions.js";
|
3
|
+
import { ApiConfiguration } from "../../../core/api/ApiConfiguration.js";
|
4
|
+
import { ResponseHandler } from "../../../core/api/postToApi.js";
|
5
|
+
import { AbstractModel } from "../../../model-function/AbstractModel.js";
|
6
|
+
import { Delta } from "../../../model-function/Delta.js";
|
7
|
+
import { TextGenerationModelSettings } from "../../../model-function/generate-text/TextGenerationModel.js";
|
8
|
+
import { ToolDefinition } from "../../../tool/ToolDefinition.js";
|
9
|
+
import { OpenAIChatMessage } from "./OpenAIChatMessage.js";
|
10
|
+
export interface AbstractOpenAIChatCallSettings {
|
11
|
+
api?: ApiConfiguration;
|
12
|
+
model: string;
|
13
|
+
functions?: Array<{
|
14
|
+
name: string;
|
15
|
+
description?: string;
|
16
|
+
parameters: unknown;
|
17
|
+
}>;
|
18
|
+
functionCall?: "none" | "auto" | {
|
19
|
+
name: string;
|
20
|
+
};
|
21
|
+
tools?: Array<{
|
22
|
+
type: "function";
|
23
|
+
function: {
|
24
|
+
name: string;
|
25
|
+
description?: string;
|
26
|
+
parameters: unknown;
|
27
|
+
};
|
28
|
+
}>;
|
29
|
+
toolChoice?: "none" | "auto" | {
|
30
|
+
type: "function";
|
31
|
+
function: {
|
32
|
+
name: string;
|
33
|
+
};
|
34
|
+
};
|
35
|
+
stop?: string | string[];
|
36
|
+
maxTokens?: number;
|
37
|
+
temperature?: number;
|
38
|
+
topP?: number;
|
39
|
+
seed?: number | null;
|
40
|
+
responseFormat?: {
|
41
|
+
type?: "text" | "json_object";
|
42
|
+
};
|
43
|
+
n?: number;
|
44
|
+
presencePenalty?: number;
|
45
|
+
frequencyPenalty?: number;
|
46
|
+
logitBias?: Record<number, number>;
|
47
|
+
}
|
48
|
+
export interface AbstractOpenAIChatSettings extends TextGenerationModelSettings, Omit<AbstractOpenAIChatCallSettings, "stop" | "maxTokens"> {
|
49
|
+
isUserIdForwardingEnabled?: boolean;
|
50
|
+
}
|
51
|
+
export type OpenAIChatPrompt = OpenAIChatMessage[];
|
52
|
+
/**
|
53
|
+
* Abstract text generation model that calls an API that is compatible with the OpenAI chat API.
|
54
|
+
*
|
55
|
+
* @see https://platform.openai.com/docs/api-reference/chat/create
|
56
|
+
*/
|
57
|
+
export declare abstract class AbstractOpenAIChatModel<SETTINGS extends AbstractOpenAIChatSettings> extends AbstractModel<SETTINGS> {
|
58
|
+
constructor(settings: SETTINGS);
|
59
|
+
callAPI<RESULT>(messages: Array<OpenAIChatMessage>, options: {
|
60
|
+
responseFormat: OpenAIChatResponseFormatType<RESULT>;
|
61
|
+
} & FunctionOptions & {
|
62
|
+
functions?: AbstractOpenAIChatCallSettings["functions"];
|
63
|
+
functionCall?: AbstractOpenAIChatCallSettings["functionCall"];
|
64
|
+
tools?: AbstractOpenAIChatCallSettings["tools"];
|
65
|
+
toolChoice?: AbstractOpenAIChatCallSettings["toolChoice"];
|
66
|
+
}): Promise<RESULT>;
|
67
|
+
doGenerateText(prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<{
|
68
|
+
response: {
|
69
|
+
object: "chat.completion";
|
70
|
+
usage: {
|
71
|
+
prompt_tokens: number;
|
72
|
+
total_tokens: number;
|
73
|
+
completion_tokens: number;
|
74
|
+
};
|
75
|
+
model: string;
|
76
|
+
id: string;
|
77
|
+
choices: {
|
78
|
+
message: {
|
79
|
+
role: "assistant";
|
80
|
+
content: string | null;
|
81
|
+
function_call?: {
|
82
|
+
name: string;
|
83
|
+
arguments: string;
|
84
|
+
} | undefined;
|
85
|
+
tool_calls?: {
|
86
|
+
function: {
|
87
|
+
name: string;
|
88
|
+
arguments: string;
|
89
|
+
};
|
90
|
+
type: "function";
|
91
|
+
id: string;
|
92
|
+
}[] | undefined;
|
93
|
+
};
|
94
|
+
index: number;
|
95
|
+
logprobs?: any;
|
96
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
97
|
+
}[];
|
98
|
+
created: number;
|
99
|
+
system_fingerprint?: string | undefined;
|
100
|
+
};
|
101
|
+
text: string;
|
102
|
+
usage: {
|
103
|
+
promptTokens: number;
|
104
|
+
completionTokens: number;
|
105
|
+
totalTokens: number;
|
106
|
+
};
|
107
|
+
}>;
|
108
|
+
doStreamText(prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<AsyncIterable<Delta<string>>>;
|
109
|
+
doGenerateToolCall(tool: ToolDefinition<string, unknown>, prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<{
|
110
|
+
response: {
|
111
|
+
object: "chat.completion";
|
112
|
+
usage: {
|
113
|
+
prompt_tokens: number;
|
114
|
+
total_tokens: number;
|
115
|
+
completion_tokens: number;
|
116
|
+
};
|
117
|
+
model: string;
|
118
|
+
id: string;
|
119
|
+
choices: {
|
120
|
+
message: {
|
121
|
+
role: "assistant";
|
122
|
+
content: string | null;
|
123
|
+
function_call?: {
|
124
|
+
name: string;
|
125
|
+
arguments: string;
|
126
|
+
} | undefined;
|
127
|
+
tool_calls?: {
|
128
|
+
function: {
|
129
|
+
name: string;
|
130
|
+
arguments: string;
|
131
|
+
};
|
132
|
+
type: "function";
|
133
|
+
id: string;
|
134
|
+
}[] | undefined;
|
135
|
+
};
|
136
|
+
index: number;
|
137
|
+
logprobs?: any;
|
138
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
139
|
+
}[];
|
140
|
+
created: number;
|
141
|
+
system_fingerprint?: string | undefined;
|
142
|
+
};
|
143
|
+
toolCall: {
|
144
|
+
id: string;
|
145
|
+
args: unknown;
|
146
|
+
} | null;
|
147
|
+
usage: {
|
148
|
+
promptTokens: number;
|
149
|
+
completionTokens: number;
|
150
|
+
totalTokens: number;
|
151
|
+
};
|
152
|
+
}>;
|
153
|
+
doGenerateToolCallsOrText(tools: Array<ToolDefinition<string, unknown>>, prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<{
|
154
|
+
response: {
|
155
|
+
object: "chat.completion";
|
156
|
+
usage: {
|
157
|
+
prompt_tokens: number;
|
158
|
+
total_tokens: number;
|
159
|
+
completion_tokens: number;
|
160
|
+
};
|
161
|
+
model: string;
|
162
|
+
id: string;
|
163
|
+
choices: {
|
164
|
+
message: {
|
165
|
+
role: "assistant";
|
166
|
+
content: string | null;
|
167
|
+
function_call?: {
|
168
|
+
name: string;
|
169
|
+
arguments: string;
|
170
|
+
} | undefined;
|
171
|
+
tool_calls?: {
|
172
|
+
function: {
|
173
|
+
name: string;
|
174
|
+
arguments: string;
|
175
|
+
};
|
176
|
+
type: "function";
|
177
|
+
id: string;
|
178
|
+
}[] | undefined;
|
179
|
+
};
|
180
|
+
index: number;
|
181
|
+
logprobs?: any;
|
182
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
183
|
+
}[];
|
184
|
+
created: number;
|
185
|
+
system_fingerprint?: string | undefined;
|
186
|
+
};
|
187
|
+
text: string | null;
|
188
|
+
toolCalls: {
|
189
|
+
id: string;
|
190
|
+
name: string;
|
191
|
+
args: unknown;
|
192
|
+
}[] | null;
|
193
|
+
usage: {
|
194
|
+
promptTokens: number;
|
195
|
+
completionTokens: number;
|
196
|
+
totalTokens: number;
|
197
|
+
};
|
198
|
+
}>;
|
199
|
+
extractUsage(response: OpenAIChatResponse): {
|
200
|
+
promptTokens: number;
|
201
|
+
completionTokens: number;
|
202
|
+
totalTokens: number;
|
203
|
+
};
|
204
|
+
}
|
205
|
+
declare const openAIChatResponseSchema: z.ZodObject<{
|
206
|
+
id: z.ZodString;
|
207
|
+
choices: z.ZodArray<z.ZodObject<{
|
208
|
+
message: z.ZodObject<{
|
209
|
+
role: z.ZodLiteral<"assistant">;
|
210
|
+
content: z.ZodNullable<z.ZodString>;
|
211
|
+
function_call: z.ZodOptional<z.ZodObject<{
|
212
|
+
name: z.ZodString;
|
213
|
+
arguments: z.ZodString;
|
214
|
+
}, "strip", z.ZodTypeAny, {
|
215
|
+
name: string;
|
216
|
+
arguments: string;
|
217
|
+
}, {
|
218
|
+
name: string;
|
219
|
+
arguments: string;
|
220
|
+
}>>;
|
221
|
+
tool_calls: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
222
|
+
id: z.ZodString;
|
223
|
+
type: z.ZodLiteral<"function">;
|
224
|
+
function: z.ZodObject<{
|
225
|
+
name: z.ZodString;
|
226
|
+
arguments: z.ZodString;
|
227
|
+
}, "strip", z.ZodTypeAny, {
|
228
|
+
name: string;
|
229
|
+
arguments: string;
|
230
|
+
}, {
|
231
|
+
name: string;
|
232
|
+
arguments: string;
|
233
|
+
}>;
|
234
|
+
}, "strip", z.ZodTypeAny, {
|
235
|
+
function: {
|
236
|
+
name: string;
|
237
|
+
arguments: string;
|
238
|
+
};
|
239
|
+
type: "function";
|
240
|
+
id: string;
|
241
|
+
}, {
|
242
|
+
function: {
|
243
|
+
name: string;
|
244
|
+
arguments: string;
|
245
|
+
};
|
246
|
+
type: "function";
|
247
|
+
id: string;
|
248
|
+
}>, "many">>;
|
249
|
+
}, "strip", z.ZodTypeAny, {
|
250
|
+
role: "assistant";
|
251
|
+
content: string | null;
|
252
|
+
function_call?: {
|
253
|
+
name: string;
|
254
|
+
arguments: string;
|
255
|
+
} | undefined;
|
256
|
+
tool_calls?: {
|
257
|
+
function: {
|
258
|
+
name: string;
|
259
|
+
arguments: string;
|
260
|
+
};
|
261
|
+
type: "function";
|
262
|
+
id: string;
|
263
|
+
}[] | undefined;
|
264
|
+
}, {
|
265
|
+
role: "assistant";
|
266
|
+
content: string | null;
|
267
|
+
function_call?: {
|
268
|
+
name: string;
|
269
|
+
arguments: string;
|
270
|
+
} | undefined;
|
271
|
+
tool_calls?: {
|
272
|
+
function: {
|
273
|
+
name: string;
|
274
|
+
arguments: string;
|
275
|
+
};
|
276
|
+
type: "function";
|
277
|
+
id: string;
|
278
|
+
}[] | undefined;
|
279
|
+
}>;
|
280
|
+
index: z.ZodNumber;
|
281
|
+
logprobs: z.ZodNullable<z.ZodAny>;
|
282
|
+
finish_reason: z.ZodNullable<z.ZodOptional<z.ZodEnum<["stop", "length", "tool_calls", "content_filter", "function_call"]>>>;
|
283
|
+
}, "strip", z.ZodTypeAny, {
|
284
|
+
message: {
|
285
|
+
role: "assistant";
|
286
|
+
content: string | null;
|
287
|
+
function_call?: {
|
288
|
+
name: string;
|
289
|
+
arguments: string;
|
290
|
+
} | undefined;
|
291
|
+
tool_calls?: {
|
292
|
+
function: {
|
293
|
+
name: string;
|
294
|
+
arguments: string;
|
295
|
+
};
|
296
|
+
type: "function";
|
297
|
+
id: string;
|
298
|
+
}[] | undefined;
|
299
|
+
};
|
300
|
+
index: number;
|
301
|
+
logprobs?: any;
|
302
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
303
|
+
}, {
|
304
|
+
message: {
|
305
|
+
role: "assistant";
|
306
|
+
content: string | null;
|
307
|
+
function_call?: {
|
308
|
+
name: string;
|
309
|
+
arguments: string;
|
310
|
+
} | undefined;
|
311
|
+
tool_calls?: {
|
312
|
+
function: {
|
313
|
+
name: string;
|
314
|
+
arguments: string;
|
315
|
+
};
|
316
|
+
type: "function";
|
317
|
+
id: string;
|
318
|
+
}[] | undefined;
|
319
|
+
};
|
320
|
+
index: number;
|
321
|
+
logprobs?: any;
|
322
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
323
|
+
}>, "many">;
|
324
|
+
created: z.ZodNumber;
|
325
|
+
model: z.ZodString;
|
326
|
+
system_fingerprint: z.ZodOptional<z.ZodString>;
|
327
|
+
object: z.ZodLiteral<"chat.completion">;
|
328
|
+
usage: z.ZodObject<{
|
329
|
+
prompt_tokens: z.ZodNumber;
|
330
|
+
completion_tokens: z.ZodNumber;
|
331
|
+
total_tokens: z.ZodNumber;
|
332
|
+
}, "strip", z.ZodTypeAny, {
|
333
|
+
prompt_tokens: number;
|
334
|
+
total_tokens: number;
|
335
|
+
completion_tokens: number;
|
336
|
+
}, {
|
337
|
+
prompt_tokens: number;
|
338
|
+
total_tokens: number;
|
339
|
+
completion_tokens: number;
|
340
|
+
}>;
|
341
|
+
}, "strip", z.ZodTypeAny, {
|
342
|
+
object: "chat.completion";
|
343
|
+
usage: {
|
344
|
+
prompt_tokens: number;
|
345
|
+
total_tokens: number;
|
346
|
+
completion_tokens: number;
|
347
|
+
};
|
348
|
+
model: string;
|
349
|
+
id: string;
|
350
|
+
choices: {
|
351
|
+
message: {
|
352
|
+
role: "assistant";
|
353
|
+
content: string | null;
|
354
|
+
function_call?: {
|
355
|
+
name: string;
|
356
|
+
arguments: string;
|
357
|
+
} | undefined;
|
358
|
+
tool_calls?: {
|
359
|
+
function: {
|
360
|
+
name: string;
|
361
|
+
arguments: string;
|
362
|
+
};
|
363
|
+
type: "function";
|
364
|
+
id: string;
|
365
|
+
}[] | undefined;
|
366
|
+
};
|
367
|
+
index: number;
|
368
|
+
logprobs?: any;
|
369
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
370
|
+
}[];
|
371
|
+
created: number;
|
372
|
+
system_fingerprint?: string | undefined;
|
373
|
+
}, {
|
374
|
+
object: "chat.completion";
|
375
|
+
usage: {
|
376
|
+
prompt_tokens: number;
|
377
|
+
total_tokens: number;
|
378
|
+
completion_tokens: number;
|
379
|
+
};
|
380
|
+
model: string;
|
381
|
+
id: string;
|
382
|
+
choices: {
|
383
|
+
message: {
|
384
|
+
role: "assistant";
|
385
|
+
content: string | null;
|
386
|
+
function_call?: {
|
387
|
+
name: string;
|
388
|
+
arguments: string;
|
389
|
+
} | undefined;
|
390
|
+
tool_calls?: {
|
391
|
+
function: {
|
392
|
+
name: string;
|
393
|
+
arguments: string;
|
394
|
+
};
|
395
|
+
type: "function";
|
396
|
+
id: string;
|
397
|
+
}[] | undefined;
|
398
|
+
};
|
399
|
+
index: number;
|
400
|
+
logprobs?: any;
|
401
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
402
|
+
}[];
|
403
|
+
created: number;
|
404
|
+
system_fingerprint?: string | undefined;
|
405
|
+
}>;
|
406
|
+
export type OpenAIChatResponse = z.infer<typeof openAIChatResponseSchema>;
|
407
|
+
export type OpenAIChatResponseFormatType<T> = {
|
408
|
+
stream: boolean;
|
409
|
+
handler: ResponseHandler<T>;
|
410
|
+
};
|
411
|
+
export declare const OpenAIChatResponseFormat: {
|
412
|
+
/**
|
413
|
+
* Returns the response as a JSON object.
|
414
|
+
*/
|
415
|
+
json: {
|
416
|
+
stream: false;
|
417
|
+
handler: ResponseHandler<{
|
418
|
+
object: "chat.completion";
|
419
|
+
usage: {
|
420
|
+
prompt_tokens: number;
|
421
|
+
total_tokens: number;
|
422
|
+
completion_tokens: number;
|
423
|
+
};
|
424
|
+
model: string;
|
425
|
+
id: string;
|
426
|
+
choices: {
|
427
|
+
message: {
|
428
|
+
role: "assistant";
|
429
|
+
content: string | null;
|
430
|
+
function_call?: {
|
431
|
+
name: string;
|
432
|
+
arguments: string;
|
433
|
+
} | undefined;
|
434
|
+
tool_calls?: {
|
435
|
+
function: {
|
436
|
+
name: string;
|
437
|
+
arguments: string;
|
438
|
+
};
|
439
|
+
type: "function";
|
440
|
+
id: string;
|
441
|
+
}[] | undefined;
|
442
|
+
};
|
443
|
+
index: number;
|
444
|
+
logprobs?: any;
|
445
|
+
finish_reason?: "length" | "stop" | "tool_calls" | "function_call" | "content_filter" | null | undefined;
|
446
|
+
}[];
|
447
|
+
created: number;
|
448
|
+
system_fingerprint?: string | undefined;
|
449
|
+
}>;
|
450
|
+
};
|
451
|
+
/**
|
452
|
+
* Returns an async iterable over the text deltas (only the tex different of the first choice).
|
453
|
+
*/
|
454
|
+
textDeltaIterable: {
|
455
|
+
stream: true;
|
456
|
+
handler: ({ response }: {
|
457
|
+
response: Response;
|
458
|
+
}) => Promise<AsyncIterable<Delta<string>>>;
|
459
|
+
};
|
460
|
+
structureDeltaIterable: {
|
461
|
+
stream: true;
|
462
|
+
handler: ({ response }: {
|
463
|
+
response: Response;
|
464
|
+
}) => Promise<AsyncIterable<Delta<unknown>>>;
|
465
|
+
};
|
466
|
+
};
|
467
|
+
export {};
|
@@ -0,0 +1,224 @@
|
|
1
|
+
import { z } from "zod";
|
2
|
+
import { callWithRetryAndThrottle } from "../../../core/api/callWithRetryAndThrottle.js";
|
3
|
+
import { createJsonResponseHandler, postJsonToApi, } from "../../../core/api/postToApi.js";
|
4
|
+
import { parseJSON } from "../../../core/schema/parseJSON.js";
|
5
|
+
import { AbstractModel } from "../../../model-function/AbstractModel.js";
|
6
|
+
import { parsePartialJson } from "../../../model-function/generate-structure/parsePartialJson.js";
|
7
|
+
import { OpenAIApiConfiguration } from "../OpenAIApiConfiguration.js";
|
8
|
+
import { failedOpenAICallResponseHandler } from "../OpenAIError.js";
|
9
|
+
import { createOpenAIChatDeltaIterableQueue } from "./OpenAIChatStreamIterable.js";
|
10
|
+
/**
|
11
|
+
* Abstract text generation model that calls an API that is compatible with the OpenAI chat API.
|
12
|
+
*
|
13
|
+
* @see https://platform.openai.com/docs/api-reference/chat/create
|
14
|
+
*/
|
15
|
+
export class AbstractOpenAIChatModel extends AbstractModel {
|
16
|
+
constructor(settings) {
|
17
|
+
super({ settings });
|
18
|
+
}
|
19
|
+
async callAPI(messages, options) {
|
20
|
+
return callWithRetryAndThrottle({
|
21
|
+
retry: this.settings.api?.retry,
|
22
|
+
throttle: this.settings.api?.throttle,
|
23
|
+
call: async () => callOpenAIChatCompletionAPI({
|
24
|
+
...this.settings,
|
25
|
+
// function & tool calling:
|
26
|
+
functions: options.functions ?? this.settings.functions,
|
27
|
+
functionCall: options.functionCall ?? this.settings.functionCall,
|
28
|
+
tools: options.tools ?? this.settings.tools,
|
29
|
+
toolChoice: options.toolChoice ?? this.settings.toolChoice,
|
30
|
+
// map to OpenAI API names:
|
31
|
+
stop: this.settings.stopSequences,
|
32
|
+
maxTokens: this.settings.maxCompletionTokens,
|
33
|
+
openAIResponseFormat: this.settings.responseFormat,
|
34
|
+
// other settings:
|
35
|
+
user: this.settings.isUserIdForwardingEnabled
|
36
|
+
? options.run?.userId
|
37
|
+
: undefined,
|
38
|
+
abortSignal: options.run?.abortSignal,
|
39
|
+
responseFormat: options.responseFormat,
|
40
|
+
messages,
|
41
|
+
}),
|
42
|
+
});
|
43
|
+
}
|
44
|
+
async doGenerateText(prompt, options) {
|
45
|
+
const response = await this.callAPI(prompt, {
|
46
|
+
...options,
|
47
|
+
responseFormat: OpenAIChatResponseFormat.json,
|
48
|
+
});
|
49
|
+
return {
|
50
|
+
response,
|
51
|
+
text: response.choices[0].message.content,
|
52
|
+
usage: this.extractUsage(response),
|
53
|
+
};
|
54
|
+
}
|
55
|
+
doStreamText(prompt, options) {
|
56
|
+
return this.callAPI(prompt, {
|
57
|
+
...options,
|
58
|
+
responseFormat: OpenAIChatResponseFormat.textDeltaIterable,
|
59
|
+
});
|
60
|
+
}
|
61
|
+
async doGenerateToolCall(tool, prompt, options) {
|
62
|
+
const response = await this.callAPI(prompt, {
|
63
|
+
...options,
|
64
|
+
responseFormat: OpenAIChatResponseFormat.json,
|
65
|
+
toolChoice: {
|
66
|
+
type: "function",
|
67
|
+
function: { name: tool.name },
|
68
|
+
},
|
69
|
+
tools: [
|
70
|
+
{
|
71
|
+
type: "function",
|
72
|
+
function: {
|
73
|
+
name: tool.name,
|
74
|
+
description: tool.description,
|
75
|
+
parameters: tool.parameters.getJsonSchema(),
|
76
|
+
},
|
77
|
+
},
|
78
|
+
],
|
79
|
+
});
|
80
|
+
const toolCalls = response.choices[0]?.message.tool_calls;
|
81
|
+
return {
|
82
|
+
response,
|
83
|
+
toolCall: toolCalls == null || toolCalls.length === 0
|
84
|
+
? null
|
85
|
+
: {
|
86
|
+
id: toolCalls[0].id,
|
87
|
+
args: parseJSON({ text: toolCalls[0].function.arguments }),
|
88
|
+
},
|
89
|
+
usage: this.extractUsage(response),
|
90
|
+
};
|
91
|
+
}
|
92
|
+
async doGenerateToolCallsOrText(tools, prompt, options) {
|
93
|
+
const response = await this.callAPI(prompt, {
|
94
|
+
...options,
|
95
|
+
responseFormat: OpenAIChatResponseFormat.json,
|
96
|
+
toolChoice: "auto",
|
97
|
+
tools: tools.map((tool) => ({
|
98
|
+
type: "function",
|
99
|
+
function: {
|
100
|
+
name: tool.name,
|
101
|
+
description: tool.description,
|
102
|
+
parameters: tool.parameters.getJsonSchema(),
|
103
|
+
},
|
104
|
+
})),
|
105
|
+
});
|
106
|
+
const message = response.choices[0]?.message;
|
107
|
+
return {
|
108
|
+
response,
|
109
|
+
text: message.content ?? null,
|
110
|
+
toolCalls: message.tool_calls?.map((toolCall) => ({
|
111
|
+
id: toolCall.id,
|
112
|
+
name: toolCall.function.name,
|
113
|
+
args: parseJSON({ text: toolCall.function.arguments }),
|
114
|
+
})) ?? null,
|
115
|
+
usage: this.extractUsage(response),
|
116
|
+
};
|
117
|
+
}
|
118
|
+
extractUsage(response) {
|
119
|
+
return {
|
120
|
+
promptTokens: response.usage.prompt_tokens,
|
121
|
+
completionTokens: response.usage.completion_tokens,
|
122
|
+
totalTokens: response.usage.total_tokens,
|
123
|
+
};
|
124
|
+
}
|
125
|
+
}
|
126
|
+
const openAIChatResponseSchema = z.object({
|
127
|
+
id: z.string(),
|
128
|
+
choices: z.array(z.object({
|
129
|
+
message: z.object({
|
130
|
+
role: z.literal("assistant"),
|
131
|
+
content: z.string().nullable(),
|
132
|
+
function_call: z
|
133
|
+
.object({
|
134
|
+
name: z.string(),
|
135
|
+
arguments: z.string(),
|
136
|
+
})
|
137
|
+
.optional(),
|
138
|
+
tool_calls: z
|
139
|
+
.array(z.object({
|
140
|
+
id: z.string(),
|
141
|
+
type: z.literal("function"),
|
142
|
+
function: z.object({
|
143
|
+
name: z.string(),
|
144
|
+
arguments: z.string(),
|
145
|
+
}),
|
146
|
+
}))
|
147
|
+
.optional(),
|
148
|
+
}),
|
149
|
+
index: z.number(),
|
150
|
+
logprobs: z.nullable(z.any()),
|
151
|
+
finish_reason: z
|
152
|
+
.enum([
|
153
|
+
"stop",
|
154
|
+
"length",
|
155
|
+
"tool_calls",
|
156
|
+
"content_filter",
|
157
|
+
"function_call",
|
158
|
+
])
|
159
|
+
.optional()
|
160
|
+
.nullable(),
|
161
|
+
})),
|
162
|
+
created: z.number(),
|
163
|
+
model: z.string(),
|
164
|
+
system_fingerprint: z.string().optional(),
|
165
|
+
object: z.literal("chat.completion"),
|
166
|
+
usage: z.object({
|
167
|
+
prompt_tokens: z.number(),
|
168
|
+
completion_tokens: z.number(),
|
169
|
+
total_tokens: z.number(),
|
170
|
+
}),
|
171
|
+
});
|
172
|
+
async function callOpenAIChatCompletionAPI({ api = new OpenAIApiConfiguration(), abortSignal, responseFormat, model, messages, functions, functionCall, tools, toolChoice, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, logitBias, user, openAIResponseFormat, seed, }) {
|
173
|
+
// empty arrays are not allowed for stop:
|
174
|
+
if (stop != null && Array.isArray(stop) && stop.length === 0) {
|
175
|
+
stop = undefined;
|
176
|
+
}
|
177
|
+
return postJsonToApi({
|
178
|
+
url: api.assembleUrl("/chat/completions"),
|
179
|
+
headers: api.headers,
|
180
|
+
body: {
|
181
|
+
stream: responseFormat.stream,
|
182
|
+
model,
|
183
|
+
messages,
|
184
|
+
functions,
|
185
|
+
function_call: functionCall,
|
186
|
+
tools,
|
187
|
+
tool_choice: toolChoice,
|
188
|
+
temperature,
|
189
|
+
top_p: topP,
|
190
|
+
n,
|
191
|
+
stop,
|
192
|
+
max_tokens: maxTokens,
|
193
|
+
presence_penalty: presencePenalty,
|
194
|
+
frequency_penalty: frequencyPenalty,
|
195
|
+
logit_bias: logitBias,
|
196
|
+
seed,
|
197
|
+
response_format: openAIResponseFormat,
|
198
|
+
user,
|
199
|
+
},
|
200
|
+
failedResponseHandler: failedOpenAICallResponseHandler,
|
201
|
+
successfulResponseHandler: responseFormat.handler,
|
202
|
+
abortSignal,
|
203
|
+
});
|
204
|
+
}
|
205
|
+
export const OpenAIChatResponseFormat = {
|
206
|
+
/**
|
207
|
+
* Returns the response as a JSON object.
|
208
|
+
*/
|
209
|
+
json: {
|
210
|
+
stream: false,
|
211
|
+
handler: createJsonResponseHandler(openAIChatResponseSchema),
|
212
|
+
},
|
213
|
+
/**
|
214
|
+
* Returns an async iterable over the text deltas (only the tex different of the first choice).
|
215
|
+
*/
|
216
|
+
textDeltaIterable: {
|
217
|
+
stream: true,
|
218
|
+
handler: async ({ response }) => createOpenAIChatDeltaIterableQueue(response.body, (delta) => delta[0]?.delta.content ?? ""),
|
219
|
+
},
|
220
|
+
structureDeltaIterable: {
|
221
|
+
stream: true,
|
222
|
+
handler: async ({ response }) => createOpenAIChatDeltaIterableQueue(response.body, (delta) => parsePartialJson(delta[0]?.function_call?.arguments)),
|
223
|
+
},
|
224
|
+
};
|