File size: 11,652 Bytes
494c9e4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import URLHandler from '../utils/URLHandler';
import * as completionResultCache from '../utils/completionResultCache';
import type { TokenWithOffset } from './generatedSchemas';

/** 与 server.yaml basePath `/api` + `/v1/completions` 一致 */
const COMPLETIONS_PATH = '/api/v1/completions';
const COMPLETIONS_PROMPT_PATH = '/api/v1/completions/prompt';
const COMPLETIONS_STOP_PATH = '/api/v1/completions/stop';

/** 与 server_openai_definitions.yaml OpenAICompletionsRequest 对齐的最小类型 */
export type OpenAICompletionsRequest = {
    model: string;
    prompt: string;
    max_tokens?: number;
    temperature?: number;
    top_p?: number;
    stop?: string | string[];
    [key: string]: unknown;
};

export type OpenAICompletionChoice = {
    text?: string;
    index?: number;
    finish_reason?: string | null;
};

/** 与 server_openai_definitions InfoRadarCompletionPayload 对齐 */
export type InfoRadarCompletionPayload = {
    bpe_strings: TokenWithOffset[];
};

/** 与 OpenAICompletionsResponse 对齐(SSE 末条 result 的 data 与此同形) */
export type OpenAICompletionsResponse = {
    id: string;
    object: 'text_completion';
    created: number;
    model: string;
    choices: OpenAICompletionChoice[];
    usage?: {
        prompt_tokens?: number;
        completion_tokens?: number;
        total_tokens?: number;
    };
    /** 续写 token 级 real_topk / pred_topk,与主站信息密度分析字段一致 */
    info_radar?: InfoRadarCompletionPayload;
};

/**
 * 单用户串行:通知后端全局停止续写(不 await,避免阻塞 UI)。
 * Chat 页 Stop 仅调用此函数而不断开 fetch,以便仍收到末条 SSE result(含 info_radar)。
 */
export function postCompletionsStop(): void {
    const url = URLHandler.basicURL() + COMPLETIONS_STOP_PATH;
    void fetch(url, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json; charset=UTF-8' },
        body: '{}'
    }).catch(() => {
        /* 忽略:Stop 与 SSE 并行,失败时生成仍可能靠墙钟或其它路径结束 */
    });
}

export type PostCompletionsPromptOptions = {
    signal?: AbortSignal;
};

/**
 * POST /v1/completions/prompt:将用户原文套用 chat template,返回实际送入续写的完整 prompt。
 */
export async function postCompletionsPrompt(
    body: { model: string; prompt: string; system?: string },
    options: PostCompletionsPromptOptions = {}
): Promise<{ prompt_used: string }> {
    const { signal } = options;
    const url = URLHandler.basicURL() + COMPLETIONS_PROMPT_PATH;
    const payload: { model: string; prompt: string; system?: string } = {
        model: body.model,
        prompt: body.prompt
    };
    if (body.system !== undefined) {
        payload.system = body.system;
    }
    const res = await fetch(url, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json; charset=UTF-8' },
        body: JSON.stringify(payload),
        signal
    });
    const text = await res.text();
    let parsed: { success?: boolean; message?: string; prompt_used?: string };
    try {
        parsed = JSON.parse(text) as typeof parsed;
    } catch {
        throw new Error(`POST ${COMPLETIONS_PROMPT_PATH} failed: ${res.status} ${text.slice(0, 500)}`);
    }
    if (!res.ok) {
        const msg =
            typeof parsed.message === 'string' ? parsed.message : text.slice(0, 500);
        throw new Error(msg || `POST ${COMPLETIONS_PROMPT_PATH} failed: ${res.status}`);
    }
    if (typeof parsed.prompt_used !== 'string' || !parsed.prompt_used.length) {
        throw new Error('completions/prompt response missing prompt_used');
    }
    return { prompt_used: parsed.prompt_used };
}

export type PostCompletionsOptions = {
    signal?: AbortSignal;
    /** 续写增量文本 */
    onDelta?: (text: string, streamEnd: boolean) => void;
    /** 与请求体 `prompt` 一致(prompt_used);仅 Chat 等需缓存时传入 */
    cacheKey?: completionResultCache.CompletionResultCacheKey;
    /** 为 true 时跳过命中本地缓存,与 Chat 页「Force retry」一致 */
    forceRefresh?: boolean;
};

/**
 * POST /v1/completions:响应恒为 SSE(delta… → result)。
 * 末条 result 的 data 为 OpenAICompletionsResponse;可与拼接的 delta 对照校验。
 */
export async function postCompletions(
    body: OpenAICompletionsRequest,
    options: PostCompletionsOptions = {}
): Promise<OpenAICompletionsResponse> {
    const { signal, onDelta, cacheKey, forceRefresh } = options;
    const modelName = body.model;

    return new Promise((resolve, reject) => {
        let settled = false;
        let streamedText = '';

        const safeReject = (e: unknown) => {
            if (!settled) {
                settled = true;
                reject(e);
            }
        };

        const safeResolve = (v: OpenAICompletionsResponse) => {
            if (settled) return;
            if (signal?.aborted) {
                safeReject(new DOMException('Aborted', 'AbortError'));
                return;
            }
            settled = true;
            if (typeof v.choices?.[0]?.text === 'string' && cacheKey) {
                void completionResultCache.save(cacheKey, v, 'complete');
            }
            resolve(v);
        };

        const rejectIfAborted = (): boolean => {
            if (!signal?.aborted) return false;
            if (cacheKey && streamedText.length > 0) {
                void completionResultCache.save(
                    cacheKey,
                    {
                        id: `partial-${Date.now()}`,
                        object: 'text_completion',
                        created: Math.floor(Date.now() / 1000),
                        model: modelName,
                        choices: [{ text: streamedText, index: 0, finish_reason: 'abort' }],
                    },
                    'partial'
                );
            }
            safeReject(new DOMException('Aborted', 'AbortError'));
            return true;
        };

        if (cacheKey) {
            void (async () => {
                try {
                    if (forceRefresh) {
                        await completionResultCache.removeCachedEntryByContentKey(
                            completionResultCache.contentKeyForPrompt(cacheKey.prompt)
                        );
                    }
                    const cached = forceRefresh ? undefined : await completionResultCache.get(cacheKey);
                    if (cached) {
                        await completionResultCache.touch(cacheKey);
                        queueMicrotask(() => {
                            if (settled) return;
                            if (rejectIfAborted()) return;
                            const text = cached.choices?.[0]?.text;
                            if (typeof text !== 'string') {
                                safeReject(new Error('completions cache: invalid choices[0].text'));
                                return;
                            }
                            onDelta?.(text, true);
                            if (settled) return;
                            if (signal?.aborted) {
                                safeReject(new DOMException('Aborted', 'AbortError'));
                                return;
                            }
                            settled = true;
                            resolve(cached);
                        });
                        return;
                    }
                } catch (e) {
                    console.warn('[completions] read cache failed:', e);
                }
                fetchRemote();
            })();
            return;
        }

        fetchRemote();

        function fetchRemote(): void {
        fetch(URLHandler.basicURL() + COMPLETIONS_PATH, {
            method: 'POST',
            headers: { 'Content-Type': 'application/json; charset=UTF-8' },
            body: JSON.stringify(body),
            signal
        })
            .then((response) => {
                if (!response.ok) {
                    return response.text().then((t) => {
                        throw new Error(`POST ${COMPLETIONS_PATH} failed: ${response.status} ${t.slice(0, 500)}`);
                    });
                }
                const reader = response.body!.getReader();
                signal?.addEventListener('abort', () => reader.cancel(), { once: true });

                const decoder = new TextDecoder();
                let buffer = '';

                const processDataLine = (jsonStr: string) => {
                    if (settled) return;
                    if (rejectIfAborted()) return;
                    let parsed: {
                        type?: string;
                        text?: string;
                        stream_end?: boolean;
                        data?: OpenAICompletionsResponse;
                        message?: string;
                    };
                    try {
                        parsed = JSON.parse(jsonStr) as typeof parsed;
                    } catch (e) {
                        safeReject(
                            new Error(
                                `SSE event JSON parse failed: ${
                                    e instanceof SyntaxError ? e.message : String(e)
                                }`
                            )
                        );
                        return;
                    }
                    if (parsed.type === 'delta') {
                        const delta = parsed.text ?? '';
                        streamedText += delta;
                        onDelta?.(delta, Boolean(parsed.stream_end));
                    } else if (parsed.type === 'result') {
                        const data = parsed.data;
                        if (data && typeof data === 'object' && 'choices' in data) {
                            safeResolve(data as OpenAICompletionsResponse);
                        } else {
                            safeReject(new Error('completions stream: invalid result payload'));
                        }
                    } else if (parsed.type === 'error') {
                        safeReject(new Error(parsed.message || 'completions stream failed'));
                    }
                };

                const readChunk = (): Promise<void> => {
                    return reader.read().then(({ done, value }) => {
                        if (settled) return;
                        if (rejectIfAborted()) return;
                        if (done) {
                            if (buffer.trim()) {
                                const line = buffer;
                                if (line.startsWith('data: ')) processDataLine(line.slice(6));
                            }
                            if (!settled) {
                                safeReject(new Error('completions stream ended without result'));
                            }
                            return;
                        }
                        buffer += decoder.decode(value, { stream: true });
                        const lines = buffer.split('\n');
                        buffer = lines.pop() || '';
                        for (const line of lines) {
                            if (line.startsWith('data: ')) processDataLine(line.slice(6));
                        }
                        return readChunk();
                    });
                };
                return readChunk();
            })
            .catch((e) => {
                if (!settled) safeReject(e);
            });
        }
    });
}