@providerprotocol/ai 0.0.1
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/LICENSE +21 -0
- package/README.md +84 -0
- package/dist/anthropic/index.d.ts +41 -0
- package/dist/anthropic/index.js +500 -0
- package/dist/anthropic/index.js.map +1 -0
- package/dist/chunk-CUCRF5W6.js +136 -0
- package/dist/chunk-CUCRF5W6.js.map +1 -0
- package/dist/chunk-FTFX2VET.js +424 -0
- package/dist/chunk-FTFX2VET.js.map +1 -0
- package/dist/chunk-QUUX4G7U.js +117 -0
- package/dist/chunk-QUUX4G7U.js.map +1 -0
- package/dist/chunk-Y6Q7JCNP.js +39 -0
- package/dist/chunk-Y6Q7JCNP.js.map +1 -0
- package/dist/google/index.d.ts +69 -0
- package/dist/google/index.js +517 -0
- package/dist/google/index.js.map +1 -0
- package/dist/http/index.d.ts +61 -0
- package/dist/http/index.js +43 -0
- package/dist/http/index.js.map +1 -0
- package/dist/index.d.ts +792 -0
- package/dist/index.js +898 -0
- package/dist/index.js.map +1 -0
- package/dist/openai/index.d.ts +204 -0
- package/dist/openai/index.js +1340 -0
- package/dist/openai/index.js.map +1 -0
- package/dist/provider-CUJWjgNl.d.ts +192 -0
- package/dist/retry-I2661_rv.d.ts +118 -0
- package/package.json +88 -0
- package/src/anthropic/index.ts +3 -0
- package/src/core/image.ts +188 -0
- package/src/core/llm.ts +619 -0
- package/src/core/provider.ts +92 -0
- package/src/google/index.ts +3 -0
- package/src/http/errors.ts +112 -0
- package/src/http/fetch.ts +210 -0
- package/src/http/index.ts +31 -0
- package/src/http/keys.ts +136 -0
- package/src/http/retry.ts +205 -0
- package/src/http/sse.ts +136 -0
- package/src/index.ts +32 -0
- package/src/openai/index.ts +9 -0
- package/src/providers/anthropic/index.ts +17 -0
- package/src/providers/anthropic/llm.ts +196 -0
- package/src/providers/anthropic/transform.ts +452 -0
- package/src/providers/anthropic/types.ts +213 -0
- package/src/providers/google/index.ts +17 -0
- package/src/providers/google/llm.ts +203 -0
- package/src/providers/google/transform.ts +487 -0
- package/src/providers/google/types.ts +214 -0
- package/src/providers/openai/index.ts +151 -0
- package/src/providers/openai/llm.completions.ts +201 -0
- package/src/providers/openai/llm.responses.ts +211 -0
- package/src/providers/openai/transform.completions.ts +628 -0
- package/src/providers/openai/transform.responses.ts +718 -0
- package/src/providers/openai/types.ts +711 -0
- package/src/types/content.ts +133 -0
- package/src/types/errors.ts +85 -0
- package/src/types/index.ts +105 -0
- package/src/types/llm.ts +211 -0
- package/src/types/messages.ts +182 -0
- package/src/types/provider.ts +195 -0
- package/src/types/schema.ts +58 -0
- package/src/types/stream.ts +146 -0
- package/src/types/thread.ts +226 -0
- package/src/types/tool.ts +88 -0
- package/src/types/turn.ts +118 -0
- package/src/utils/id.ts +28 -0
|
@@ -0,0 +1,214 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Google Gemini-specific LLM parameters
|
|
3
|
+
*/
|
|
4
|
+
export interface GoogleLLMParams {
|
|
5
|
+
/** Maximum number of tokens to generate */
|
|
6
|
+
maxOutputTokens?: number;
|
|
7
|
+
|
|
8
|
+
/** Temperature for randomness (0.0 - 2.0) */
|
|
9
|
+
temperature?: number;
|
|
10
|
+
|
|
11
|
+
/** Top-p (nucleus) sampling */
|
|
12
|
+
topP?: number;
|
|
13
|
+
|
|
14
|
+
/** Top-k sampling */
|
|
15
|
+
topK?: number;
|
|
16
|
+
|
|
17
|
+
/** Stop sequences */
|
|
18
|
+
stopSequences?: string[];
|
|
19
|
+
|
|
20
|
+
/** Number of candidates to generate */
|
|
21
|
+
candidateCount?: number;
|
|
22
|
+
|
|
23
|
+
/** Response MIME type */
|
|
24
|
+
responseMimeType?: 'text/plain' | 'application/json';
|
|
25
|
+
|
|
26
|
+
/** Response schema for structured output */
|
|
27
|
+
responseSchema?: Record<string, unknown>;
|
|
28
|
+
|
|
29
|
+
/**
|
|
30
|
+
* Presence penalty for new topics
|
|
31
|
+
* Positive values encourage discussing new topics
|
|
32
|
+
*/
|
|
33
|
+
presencePenalty?: number;
|
|
34
|
+
|
|
35
|
+
/**
|
|
36
|
+
* Frequency penalty for repeated tokens
|
|
37
|
+
* Positive values discourage repetition
|
|
38
|
+
*/
|
|
39
|
+
frequencyPenalty?: number;
|
|
40
|
+
|
|
41
|
+
/**
|
|
42
|
+
* Seed for deterministic sampling
|
|
43
|
+
* Same seed with same parameters should produce same results
|
|
44
|
+
*/
|
|
45
|
+
seed?: number;
|
|
46
|
+
|
|
47
|
+
/**
|
|
48
|
+
* Whether to return log probabilities in response
|
|
49
|
+
*/
|
|
50
|
+
responseLogprobs?: boolean;
|
|
51
|
+
|
|
52
|
+
/**
|
|
53
|
+
* Number of log probabilities to return (requires responseLogprobs: true)
|
|
54
|
+
*/
|
|
55
|
+
logprobs?: number;
|
|
56
|
+
|
|
57
|
+
/**
|
|
58
|
+
* Whether to include audio timestamps in response
|
|
59
|
+
*/
|
|
60
|
+
audioTimestamp?: boolean;
|
|
61
|
+
|
|
62
|
+
/**
|
|
63
|
+
* Thinking/reasoning configuration for Gemini 3+ models
|
|
64
|
+
*/
|
|
65
|
+
thinkingConfig?: GoogleThinkingConfig;
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
/**
|
|
69
|
+
* Thinking configuration for Gemini 3+ models
|
|
70
|
+
*/
|
|
71
|
+
export interface GoogleThinkingConfig {
|
|
72
|
+
/** Whether thinking is enabled */
|
|
73
|
+
thinkingBudget?: number;
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
/**
|
|
77
|
+
* Google API request body
|
|
78
|
+
*/
|
|
79
|
+
export interface GoogleRequest {
|
|
80
|
+
contents: GoogleContent[];
|
|
81
|
+
systemInstruction?: {
|
|
82
|
+
parts: GooglePart[];
|
|
83
|
+
};
|
|
84
|
+
generationConfig?: {
|
|
85
|
+
maxOutputTokens?: number;
|
|
86
|
+
temperature?: number;
|
|
87
|
+
topP?: number;
|
|
88
|
+
topK?: number;
|
|
89
|
+
stopSequences?: string[];
|
|
90
|
+
candidateCount?: number;
|
|
91
|
+
responseMimeType?: string;
|
|
92
|
+
responseSchema?: Record<string, unknown>;
|
|
93
|
+
presencePenalty?: number;
|
|
94
|
+
frequencyPenalty?: number;
|
|
95
|
+
seed?: number;
|
|
96
|
+
responseLogprobs?: boolean;
|
|
97
|
+
logprobs?: number;
|
|
98
|
+
audioTimestamp?: boolean;
|
|
99
|
+
thinkingConfig?: GoogleThinkingConfig;
|
|
100
|
+
};
|
|
101
|
+
tools?: GoogleTool[];
|
|
102
|
+
safetySettings?: GoogleSafetySetting[];
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
/**
|
|
106
|
+
* Google content (message) format
|
|
107
|
+
*/
|
|
108
|
+
export interface GoogleContent {
|
|
109
|
+
role: 'user' | 'model';
|
|
110
|
+
parts: GooglePart[];
|
|
111
|
+
}
|
|
112
|
+
|
|
113
|
+
/**
|
|
114
|
+
* Google content part types
|
|
115
|
+
*/
|
|
116
|
+
export type GooglePart =
|
|
117
|
+
| GoogleTextPart
|
|
118
|
+
| GoogleImagePart
|
|
119
|
+
| GoogleFunctionCallPart
|
|
120
|
+
| GoogleFunctionResponsePart;
|
|
121
|
+
|
|
122
|
+
export interface GoogleTextPart {
|
|
123
|
+
text: string;
|
|
124
|
+
}
|
|
125
|
+
|
|
126
|
+
export interface GoogleImagePart {
|
|
127
|
+
inlineData: {
|
|
128
|
+
mimeType: string;
|
|
129
|
+
data: string;
|
|
130
|
+
};
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
export interface GoogleFunctionCallPart {
|
|
134
|
+
functionCall: {
|
|
135
|
+
name: string;
|
|
136
|
+
args: Record<string, unknown>;
|
|
137
|
+
};
|
|
138
|
+
/** Gemini 3+ thought signature for multi-turn tool calls */
|
|
139
|
+
thoughtSignature?: string;
|
|
140
|
+
}
|
|
141
|
+
|
|
142
|
+
export interface GoogleFunctionResponsePart {
|
|
143
|
+
functionResponse: {
|
|
144
|
+
name: string;
|
|
145
|
+
response: Record<string, unknown>;
|
|
146
|
+
};
|
|
147
|
+
}
|
|
148
|
+
|
|
149
|
+
/**
|
|
150
|
+
* Google tool format
|
|
151
|
+
*/
|
|
152
|
+
export interface GoogleTool {
|
|
153
|
+
functionDeclarations: GoogleFunctionDeclaration[];
|
|
154
|
+
}
|
|
155
|
+
|
|
156
|
+
export interface GoogleFunctionDeclaration {
|
|
157
|
+
name: string;
|
|
158
|
+
description: string;
|
|
159
|
+
parameters: {
|
|
160
|
+
type: 'object';
|
|
161
|
+
properties: Record<string, unknown>;
|
|
162
|
+
required?: string[];
|
|
163
|
+
};
|
|
164
|
+
}
|
|
165
|
+
|
|
166
|
+
/**
|
|
167
|
+
* Google safety setting
|
|
168
|
+
*/
|
|
169
|
+
export interface GoogleSafetySetting {
|
|
170
|
+
category: string;
|
|
171
|
+
threshold: string;
|
|
172
|
+
}
|
|
173
|
+
|
|
174
|
+
/**
|
|
175
|
+
* Google response format
|
|
176
|
+
*/
|
|
177
|
+
export interface GoogleResponse {
|
|
178
|
+
candidates: GoogleCandidate[];
|
|
179
|
+
usageMetadata?: {
|
|
180
|
+
promptTokenCount: number;
|
|
181
|
+
candidatesTokenCount: number;
|
|
182
|
+
totalTokenCount: number;
|
|
183
|
+
};
|
|
184
|
+
}
|
|
185
|
+
|
|
186
|
+
export interface GoogleCandidate {
|
|
187
|
+
content: {
|
|
188
|
+
role: 'model';
|
|
189
|
+
parts: GoogleResponsePart[];
|
|
190
|
+
};
|
|
191
|
+
finishReason: 'STOP' | 'MAX_TOKENS' | 'SAFETY' | 'RECITATION' | 'OTHER' | 'TOOL_USE' | null;
|
|
192
|
+
index: number;
|
|
193
|
+
safetyRatings?: GoogleSafetyRating[];
|
|
194
|
+
}
|
|
195
|
+
|
|
196
|
+
export type GoogleResponsePart = GoogleTextPart | GoogleFunctionCallPart;
|
|
197
|
+
|
|
198
|
+
export interface GoogleSafetyRating {
|
|
199
|
+
category: string;
|
|
200
|
+
probability: string;
|
|
201
|
+
}
|
|
202
|
+
|
|
203
|
+
/**
|
|
204
|
+
* Google streaming response chunk
|
|
205
|
+
* Same structure as regular response but may be partial
|
|
206
|
+
*/
|
|
207
|
+
export interface GoogleStreamChunk {
|
|
208
|
+
candidates?: GoogleCandidate[];
|
|
209
|
+
usageMetadata?: {
|
|
210
|
+
promptTokenCount: number;
|
|
211
|
+
candidatesTokenCount: number;
|
|
212
|
+
totalTokenCount: number;
|
|
213
|
+
};
|
|
214
|
+
}
|
|
@@ -0,0 +1,151 @@
|
|
|
1
|
+
import type {
|
|
2
|
+
Provider,
|
|
3
|
+
ModelReference,
|
|
4
|
+
LLMHandler,
|
|
5
|
+
LLMProvider,
|
|
6
|
+
} from '../../types/provider.ts';
|
|
7
|
+
import { createCompletionsLLMHandler } from './llm.completions.ts';
|
|
8
|
+
import { createResponsesLLMHandler } from './llm.responses.ts';
|
|
9
|
+
import type { OpenAILLMParams, OpenAIConfig } from './types.ts';
|
|
10
|
+
|
|
11
|
+
/**
|
|
12
|
+
* OpenAI provider options
|
|
13
|
+
*/
|
|
14
|
+
export interface OpenAIProviderOptions {
|
|
15
|
+
/**
|
|
16
|
+
* Which API to use:
|
|
17
|
+
* - 'responses': Modern Responses API (default, recommended)
|
|
18
|
+
* - 'completions': Legacy Chat Completions API
|
|
19
|
+
*/
|
|
20
|
+
api?: 'responses' | 'completions';
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
/**
|
|
24
|
+
* OpenAI provider with configurable API mode
|
|
25
|
+
*
|
|
26
|
+
* @example
|
|
27
|
+
* // Using the modern Responses API (default)
|
|
28
|
+
* const model = openai('gpt-4o');
|
|
29
|
+
*
|
|
30
|
+
* @example
|
|
31
|
+
* // Using the legacy Chat Completions API
|
|
32
|
+
* const model = openai('gpt-4o', { api: 'completions' });
|
|
33
|
+
*
|
|
34
|
+
* @example
|
|
35
|
+
* // Explicit Responses API
|
|
36
|
+
* const model = openai('gpt-4o', { api: 'responses' });
|
|
37
|
+
*/
|
|
38
|
+
export interface OpenAIProvider extends Provider<OpenAIProviderOptions> {
|
|
39
|
+
/**
|
|
40
|
+
* Create a model reference
|
|
41
|
+
* @param modelId - The model identifier (e.g., 'gpt-4o', 'gpt-4-turbo', 'o1-preview')
|
|
42
|
+
* @param options - Provider options including API selection
|
|
43
|
+
*/
|
|
44
|
+
(modelId: string, options?: OpenAIProviderOptions): ModelReference<OpenAIProviderOptions>;
|
|
45
|
+
|
|
46
|
+
/** Provider name */
|
|
47
|
+
readonly name: 'openai';
|
|
48
|
+
|
|
49
|
+
/** Provider version */
|
|
50
|
+
readonly version: string;
|
|
51
|
+
|
|
52
|
+
/** Supported modalities */
|
|
53
|
+
readonly modalities: {
|
|
54
|
+
llm: LLMHandler<OpenAILLMParams>;
|
|
55
|
+
};
|
|
56
|
+
}
|
|
57
|
+
|
|
58
|
+
/**
|
|
59
|
+
* Create the OpenAI provider
|
|
60
|
+
*/
|
|
61
|
+
function createOpenAIProvider(): OpenAIProvider {
|
|
62
|
+
// Track which API mode is currently active for the modalities
|
|
63
|
+
let currentApiMode: 'responses' | 'completions' = 'responses';
|
|
64
|
+
|
|
65
|
+
// Create handlers eagerly so we can inject provider reference
|
|
66
|
+
const responsesHandler = createResponsesLLMHandler();
|
|
67
|
+
const completionsHandler = createCompletionsLLMHandler();
|
|
68
|
+
|
|
69
|
+
const fn = function (
|
|
70
|
+
modelId: string,
|
|
71
|
+
options?: OpenAIProviderOptions
|
|
72
|
+
): ModelReference<OpenAIProviderOptions> {
|
|
73
|
+
const apiMode = options?.api ?? 'responses';
|
|
74
|
+
currentApiMode = apiMode;
|
|
75
|
+
return { modelId, provider };
|
|
76
|
+
};
|
|
77
|
+
|
|
78
|
+
// Create a dynamic modalities object that returns the correct handler
|
|
79
|
+
const modalities = {
|
|
80
|
+
get llm(): LLMHandler<OpenAILLMParams> {
|
|
81
|
+
return currentApiMode === 'completions'
|
|
82
|
+
? completionsHandler
|
|
83
|
+
: responsesHandler;
|
|
84
|
+
},
|
|
85
|
+
};
|
|
86
|
+
|
|
87
|
+
// Define properties
|
|
88
|
+
Object.defineProperties(fn, {
|
|
89
|
+
name: {
|
|
90
|
+
value: 'openai',
|
|
91
|
+
writable: false,
|
|
92
|
+
configurable: true,
|
|
93
|
+
},
|
|
94
|
+
version: {
|
|
95
|
+
value: '1.0.0',
|
|
96
|
+
writable: false,
|
|
97
|
+
configurable: true,
|
|
98
|
+
},
|
|
99
|
+
modalities: {
|
|
100
|
+
value: modalities,
|
|
101
|
+
writable: false,
|
|
102
|
+
configurable: true,
|
|
103
|
+
},
|
|
104
|
+
});
|
|
105
|
+
|
|
106
|
+
const provider = fn as OpenAIProvider;
|
|
107
|
+
|
|
108
|
+
// Inject provider reference into both handlers (spec compliance)
|
|
109
|
+
responsesHandler._setProvider?.(provider as unknown as LLMProvider<OpenAILLMParams>);
|
|
110
|
+
completionsHandler._setProvider?.(provider as unknown as LLMProvider<OpenAILLMParams>);
|
|
111
|
+
|
|
112
|
+
return provider;
|
|
113
|
+
}
|
|
114
|
+
|
|
115
|
+
/**
|
|
116
|
+
* OpenAI provider
|
|
117
|
+
*
|
|
118
|
+
* Supports both the modern Responses API (default) and legacy Chat Completions API.
|
|
119
|
+
*
|
|
120
|
+
* @example
|
|
121
|
+
* ```ts
|
|
122
|
+
* import { openai } from './providers/openai';
|
|
123
|
+
* import { llm } from './core/llm';
|
|
124
|
+
*
|
|
125
|
+
* // Using Responses API (default, modern, recommended)
|
|
126
|
+
* const model = llm({
|
|
127
|
+
* model: openai('gpt-4o'),
|
|
128
|
+
* params: { max_tokens: 1000 }
|
|
129
|
+
* });
|
|
130
|
+
*
|
|
131
|
+
* // Using Chat Completions API (legacy)
|
|
132
|
+
* const legacyModel = llm({
|
|
133
|
+
* model: openai('gpt-4o', { api: 'completions' }),
|
|
134
|
+
* params: { max_tokens: 1000 }
|
|
135
|
+
* });
|
|
136
|
+
*
|
|
137
|
+
* // Generate
|
|
138
|
+
* const turn = await model.generate('Hello!');
|
|
139
|
+
* console.log(turn.response.text);
|
|
140
|
+
* ```
|
|
141
|
+
*/
|
|
142
|
+
export const openai = createOpenAIProvider();
|
|
143
|
+
|
|
144
|
+
// Re-export types
|
|
145
|
+
export type {
|
|
146
|
+
OpenAILLMParams,
|
|
147
|
+
OpenAIConfig,
|
|
148
|
+
OpenAIAPIMode,
|
|
149
|
+
OpenAIModelOptions,
|
|
150
|
+
OpenAIModelReference,
|
|
151
|
+
} from './types.ts';
|
|
@@ -0,0 +1,201 @@
|
|
|
1
|
+
import type { LLMHandler, BoundLLMModel, LLMRequest, LLMResponse, LLMStreamResult, LLMCapabilities } from '../../types/llm.ts';
|
|
2
|
+
import type { StreamEvent } from '../../types/stream.ts';
|
|
3
|
+
import type { LLMProvider } from '../../types/provider.ts';
|
|
4
|
+
import { UPPError } from '../../types/errors.ts';
|
|
5
|
+
import { resolveApiKey } from '../../http/keys.ts';
|
|
6
|
+
import { doFetch, doStreamFetch } from '../../http/fetch.ts';
|
|
7
|
+
import { parseSSEStream } from '../../http/sse.ts';
|
|
8
|
+
import { normalizeHttpError } from '../../http/errors.ts';
|
|
9
|
+
import type { OpenAILLMParams, OpenAICompletionsResponse, OpenAICompletionsStreamChunk } from './types.ts';
|
|
10
|
+
import {
|
|
11
|
+
transformRequest,
|
|
12
|
+
transformResponse,
|
|
13
|
+
transformStreamEvent,
|
|
14
|
+
createStreamState,
|
|
15
|
+
buildResponseFromState,
|
|
16
|
+
} from './transform.completions.ts';
|
|
17
|
+
|
|
18
|
+
const OPENAI_API_URL = 'https://api.openai.com/v1/chat/completions';
|
|
19
|
+
|
|
20
|
+
/**
|
|
21
|
+
* OpenAI API capabilities
|
|
22
|
+
*/
|
|
23
|
+
const OPENAI_CAPABILITIES: LLMCapabilities = {
|
|
24
|
+
streaming: true,
|
|
25
|
+
tools: true,
|
|
26
|
+
structuredOutput: true,
|
|
27
|
+
imageInput: true,
|
|
28
|
+
videoInput: false,
|
|
29
|
+
audioInput: false,
|
|
30
|
+
};
|
|
31
|
+
|
|
32
|
+
/**
|
|
33
|
+
* Create OpenAI Chat Completions LLM handler
|
|
34
|
+
*/
|
|
35
|
+
export function createCompletionsLLMHandler(): LLMHandler<OpenAILLMParams> {
|
|
36
|
+
// Provider reference injected by createProvider() or OpenAI's custom factory
|
|
37
|
+
let providerRef: LLMProvider<OpenAILLMParams> | null = null;
|
|
38
|
+
|
|
39
|
+
return {
|
|
40
|
+
_setProvider(provider: LLMProvider<OpenAILLMParams>) {
|
|
41
|
+
providerRef = provider;
|
|
42
|
+
},
|
|
43
|
+
|
|
44
|
+
bind(modelId: string): BoundLLMModel<OpenAILLMParams> {
|
|
45
|
+
// Use the injected provider reference
|
|
46
|
+
if (!providerRef) {
|
|
47
|
+
throw new UPPError(
|
|
48
|
+
'Provider reference not set. Handler must be used with createProvider() or have _setProvider called.',
|
|
49
|
+
'INVALID_REQUEST',
|
|
50
|
+
'openai',
|
|
51
|
+
'llm'
|
|
52
|
+
);
|
|
53
|
+
}
|
|
54
|
+
|
|
55
|
+
const model: BoundLLMModel<OpenAILLMParams> = {
|
|
56
|
+
modelId,
|
|
57
|
+
capabilities: OPENAI_CAPABILITIES,
|
|
58
|
+
|
|
59
|
+
get provider(): LLMProvider<OpenAILLMParams> {
|
|
60
|
+
return providerRef!;
|
|
61
|
+
},
|
|
62
|
+
|
|
63
|
+
async complete(request: LLMRequest<OpenAILLMParams>): Promise<LLMResponse> {
|
|
64
|
+
const apiKey = await resolveApiKey(
|
|
65
|
+
request.config,
|
|
66
|
+
'OPENAI_API_KEY',
|
|
67
|
+
'openai',
|
|
68
|
+
'llm'
|
|
69
|
+
);
|
|
70
|
+
|
|
71
|
+
const baseUrl = request.config.baseUrl ?? OPENAI_API_URL;
|
|
72
|
+
const body = transformRequest(request, modelId);
|
|
73
|
+
|
|
74
|
+
const response = await doFetch(
|
|
75
|
+
baseUrl,
|
|
76
|
+
{
|
|
77
|
+
method: 'POST',
|
|
78
|
+
headers: {
|
|
79
|
+
'Content-Type': 'application/json',
|
|
80
|
+
Authorization: `Bearer ${apiKey}`,
|
|
81
|
+
},
|
|
82
|
+
body: JSON.stringify(body),
|
|
83
|
+
signal: request.signal,
|
|
84
|
+
},
|
|
85
|
+
request.config,
|
|
86
|
+
'openai',
|
|
87
|
+
'llm'
|
|
88
|
+
);
|
|
89
|
+
|
|
90
|
+
const data = (await response.json()) as OpenAICompletionsResponse;
|
|
91
|
+
return transformResponse(data);
|
|
92
|
+
},
|
|
93
|
+
|
|
94
|
+
stream(request: LLMRequest<OpenAILLMParams>): LLMStreamResult {
|
|
95
|
+
const state = createStreamState();
|
|
96
|
+
let responseResolve: (value: LLMResponse) => void;
|
|
97
|
+
let responseReject: (error: Error) => void;
|
|
98
|
+
|
|
99
|
+
const responsePromise = new Promise<LLMResponse>((resolve, reject) => {
|
|
100
|
+
responseResolve = resolve;
|
|
101
|
+
responseReject = reject;
|
|
102
|
+
});
|
|
103
|
+
|
|
104
|
+
async function* generateEvents(): AsyncGenerator<StreamEvent, void, unknown> {
|
|
105
|
+
try {
|
|
106
|
+
const apiKey = await resolveApiKey(
|
|
107
|
+
request.config,
|
|
108
|
+
'OPENAI_API_KEY',
|
|
109
|
+
'openai',
|
|
110
|
+
'llm'
|
|
111
|
+
);
|
|
112
|
+
|
|
113
|
+
const baseUrl = request.config.baseUrl ?? OPENAI_API_URL;
|
|
114
|
+
const body = transformRequest(request, modelId);
|
|
115
|
+
body.stream = true;
|
|
116
|
+
body.stream_options = { include_usage: true };
|
|
117
|
+
|
|
118
|
+
const response = await doStreamFetch(
|
|
119
|
+
baseUrl,
|
|
120
|
+
{
|
|
121
|
+
method: 'POST',
|
|
122
|
+
headers: {
|
|
123
|
+
'Content-Type': 'application/json',
|
|
124
|
+
Authorization: `Bearer ${apiKey}`,
|
|
125
|
+
},
|
|
126
|
+
body: JSON.stringify(body),
|
|
127
|
+
signal: request.signal,
|
|
128
|
+
},
|
|
129
|
+
request.config,
|
|
130
|
+
'openai',
|
|
131
|
+
'llm'
|
|
132
|
+
);
|
|
133
|
+
|
|
134
|
+
if (!response.ok) {
|
|
135
|
+
const error = await normalizeHttpError(response, 'openai', 'llm');
|
|
136
|
+
responseReject(error);
|
|
137
|
+
throw error;
|
|
138
|
+
}
|
|
139
|
+
|
|
140
|
+
if (!response.body) {
|
|
141
|
+
const error = new UPPError(
|
|
142
|
+
'No response body for streaming request',
|
|
143
|
+
'PROVIDER_ERROR',
|
|
144
|
+
'openai',
|
|
145
|
+
'llm'
|
|
146
|
+
);
|
|
147
|
+
responseReject(error);
|
|
148
|
+
throw error;
|
|
149
|
+
}
|
|
150
|
+
|
|
151
|
+
for await (const data of parseSSEStream(response.body)) {
|
|
152
|
+
// Skip [DONE] marker
|
|
153
|
+
if (data === '[DONE]') {
|
|
154
|
+
continue;
|
|
155
|
+
}
|
|
156
|
+
|
|
157
|
+
// Check for OpenAI error event
|
|
158
|
+
if (typeof data === 'object' && data !== null) {
|
|
159
|
+
const chunk = data as OpenAICompletionsStreamChunk;
|
|
160
|
+
|
|
161
|
+
// Check for error in chunk
|
|
162
|
+
if ('error' in chunk && chunk.error) {
|
|
163
|
+
const errorData = chunk.error as { message?: string; type?: string };
|
|
164
|
+
const error = new UPPError(
|
|
165
|
+
errorData.message ?? 'Unknown error',
|
|
166
|
+
'PROVIDER_ERROR',
|
|
167
|
+
'openai',
|
|
168
|
+
'llm'
|
|
169
|
+
);
|
|
170
|
+
responseReject(error);
|
|
171
|
+
throw error;
|
|
172
|
+
}
|
|
173
|
+
|
|
174
|
+
const uppEvents = transformStreamEvent(chunk, state);
|
|
175
|
+
for (const event of uppEvents) {
|
|
176
|
+
yield event;
|
|
177
|
+
}
|
|
178
|
+
}
|
|
179
|
+
}
|
|
180
|
+
|
|
181
|
+
// Build final response
|
|
182
|
+
responseResolve(buildResponseFromState(state));
|
|
183
|
+
} catch (error) {
|
|
184
|
+
responseReject(error as Error);
|
|
185
|
+
throw error;
|
|
186
|
+
}
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
return {
|
|
190
|
+
[Symbol.asyncIterator]() {
|
|
191
|
+
return generateEvents();
|
|
192
|
+
},
|
|
193
|
+
response: responsePromise,
|
|
194
|
+
};
|
|
195
|
+
},
|
|
196
|
+
};
|
|
197
|
+
|
|
198
|
+
return model;
|
|
199
|
+
},
|
|
200
|
+
};
|
|
201
|
+
}
|