X/LittleWhiteBox
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(story-summary): make vector APIs configurable

Browse Source
This commit is contained in:
bielie
2026-04-03 15:31:13 +08:00
parent 5424dae2d6
commit af7e0f689d
9 changed files with 468 additions and 187 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
modules/story-summary/vector/llm/llm-service.js
View File

@@ -2,14 +2,15 @@
// vector/llm/llm-service.js - 修复 prefill 传递方式
// ═══════════════════════════════════════════════════════════════════════════
import { xbLog } from '../../../../core/debug-core.js';
import { getApiKey } from './siliconflow.js';
import { getVectorConfig } from '../../data/config.js';
const MODULE_ID = 'vector-llm-service';
const SILICONFLOW_API_URL = 'https://api.siliconflow.cn/v1';
const DEFAULT_L0_MODEL = 'Qwen/Qwen3-8B';
const DEFAULT_L0_API_URL = 'https://api.siliconflow.cn/v1';
let callCounter = 0;
const activeL0SessionIds = new Set();
let l0KeyIndex = 0;
function getStreamingModule() {
const mod = window.xiaobaixStreamingGeneration;
@@ -28,6 +29,28 @@ function b64UrlEncode(str) {
return btoa(bin).replace(/\+/g, '-').replace(/\//g, '_').replace(/=+$/, '');
}
function getL0ApiConfig() {
const cfg = getVectorConfig() || {};
return cfg.l0Api || {
provider: 'siliconflow',
url: DEFAULT_L0_API_URL,
key: '',
model: DEFAULT_L0_MODEL,
};
}
function getNextKey(rawKey) {
const keys = String(rawKey || '')
.split(/[,;|\n]+/)
.map(k => k.trim())
.filter(Boolean);
if (!keys.length) return '';
if (keys.length === 1) return keys[0];
const idx = l0KeyIndex % keys.length;
l0KeyIndex = (l0KeyIndex + 1) % keys.length;
return keys[idx];
}
/**
* 统一LLM调用 - 走酒馆后端（非流式）
* 临时改为标准 messages 调用，避免 bottomassistant prefill 兼容性问题。
@@ -42,7 +65,8 @@ export async function callLLM(messages, options = {}) {
const mod = getStreamingModule();
if (!mod) throw new Error('Streaming module not ready');
const apiKey = getApiKey() || '';
const apiCfg = getL0ApiConfig();
const apiKey = getNextKey(apiCfg.key);
if (!apiKey) {
throw new Error('L0 requires siliconflow API key');
}
@@ -60,11 +84,11 @@ export async function callLLM(messages, options = {}) {
temperature: String(temperature),
max_tokens: String(max_tokens),
api: 'openai',
apiurl: SILICONFLOW_API_URL,
apiurl: String(apiCfg.url || DEFAULT_L0_API_URL).trim(),
apipassword: apiKey,
model: DEFAULT_L0_MODEL,
model: String(apiCfg.model || DEFAULT_L0_MODEL).trim(),
};
const isQwen3 = String(DEFAULT_L0_MODEL || '').includes('Qwen3');
const isQwen3 = String(args.model || '').includes('Qwen3');
if (isQwen3) {
args.enable_thinking = 'false';
}

35
modules/story-summary/vector/llm/reranker.js
View File

@@ -4,15 +4,38 @@
// ═══════════════════════════════════════════════════════════════════════════
import { xbLog } from '../../../../core/debug-core.js';
import { getApiKey } from './siliconflow.js';
import { getVectorConfig } from '../../data/config.js';
const MODULE_ID = 'reranker';
const RERANK_URL = 'https://api.siliconflow.cn/v1/rerank';
const DEFAULT_RERANK_URL = 'https://api.siliconflow.cn/v1';
const RERANK_MODEL = 'BAAI/bge-reranker-v2-m3';
const DEFAULT_TIMEOUT = 15000;
const MAX_DOCUMENTS = 100; // API 限制
const RERANK_BATCH_SIZE = 20;
const RERANK_MAX_CONCURRENCY = 5;
let rerankKeyIndex = 0;
function getRerankApiConfig() {
const cfg = getVectorConfig() || {};
return cfg.rerankApi || {
provider: 'siliconflow',
url: DEFAULT_RERANK_URL,
key: '',
model: RERANK_MODEL,
};
}
function getNextRerankKey(rawKey) {
const keys = String(rawKey || '')
.split(/[,;|\n]+/)
.map(k => k.trim())
.filter(Boolean);
if (!keys.length) return '';
if (keys.length === 1) return keys[0];
const idx = rerankKeyIndex % keys.length;
rerankKeyIndex = (rerankKeyIndex + 1) % keys.length;
return keys[idx];
}
/**
* 对文档列表进行 Rerank 精排
@@ -37,7 +60,8 @@ export async function rerank(query, documents, options = {}) {
return { results: [], failed: false };
}
const key = getApiKey();
const apiCfg = getRerankApiConfig();
const key = getNextRerankKey(apiCfg.key);
if (!key) {
xbLog.warn(MODULE_ID, '未配置 API Key，跳过 rerank');
return { results: documents.map((_, i) => ({ index: i, relevance_score: 0 })), failed: true };
@@ -72,14 +96,15 @@ export async function rerank(query, documents, options = {}) {
try {
const T0 = performance.now();
const response = await fetch(RERANK_URL, {
const baseUrl = String(apiCfg.url || DEFAULT_RERANK_URL).replace(/\/+$/, '');
const response = await fetch(`${baseUrl}/rerank`, {
method: 'POST',
headers: {
'Authorization': `Bearer ${key}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
model: RERANK_MODEL,
model: String(apiCfg.model || RERANK_MODEL),
// Zero-darkbox: do not silently truncate query.
query,
documents: validDocs,

46
modules/story-summary/vector/llm/siliconflow.js
View File

@@ -1,31 +1,39 @@
// ═══════════════════════════════════════════════════════════════════════════
// siliconflow.js - Embedding + 多 Key 轮询
// siliconflow.js - OpenAI-compatible Embedding + 多 Key 轮询
//
// 在 API Key 输入框中用逗号、分号、竖线或换行分隔多个 Key，例如：
// sk-aaa,sk-bbb,sk-ccc
// 每次调用自动轮询到下一个 Key，并发请求会均匀分布到所有 Key 上。
// ═══════════════════════════════════════════════════════════════════════════
import { getVectorConfig } from '../../data/config.js';
const BASE_URL = 'https://api.siliconflow.cn';
const EMBEDDING_MODEL = 'BAAI/bge-m3';
// ★ 多 Key 轮询状态
let _keyIndex = 0;
function getEmbeddingApiConfig() {
const cfg = getVectorConfig() || {};
return cfg.embeddingApi || {
provider: 'siliconflow',
url: `${BASE_URL}/v1`,
key: '',
model: EMBEDDING_MODEL,
};
}
/**
* 从 localStorage 解析所有 Key（支持逗号、分号、竖线、换行分隔）
*/
function parseKeys() {
function parseKeys(rawKey) {
try {
const raw = localStorage.getItem('summary_panel_config');
if (raw) {
const parsed = JSON.parse(raw);
const keyStr = parsed.vector?.online?.key || '';
return keyStr
.split(/[,;|\n]+/)
.map(k => k.trim())
.filter(k => k.length > 0);
}
const keyStr = String(rawKey || '');
return keyStr
.split(/[,;|\n]+/)
.map(k => k.trim())
.filter(k => k.length > 0);
} catch { }
return [];
}
@@ -34,8 +42,8 @@ function parseKeys() {
* 获取下一个可用的 API Key（轮询）
* 每次调用返回不同的 Key，自动循环
*/
export function getApiKey() {
const keys = parseKeys();
export function getApiKey(rawKey = null) {
const keys = parseKeys(rawKey ?? getEmbeddingApiConfig().key);
if (!keys.length) return null;
if (keys.length === 1) return keys[0];
@@ -51,7 +59,7 @@ export function getApiKey() {
* 获取当前配置的 Key 数量（供外部模块动态调整并发用）
*/
export function getKeyCount() {
return Math.max(1, parseKeys().length);
return Math.max(1, parseKeys(getEmbeddingApiConfig().key).length);
}
// ═══════════════════════════════════════════════════════════════════════════
@@ -61,22 +69,24 @@ export function getKeyCount() {
export async function embed(texts, options = {}) {
if (!texts?.length) return [];
const key = getApiKey();
if (!key) throw new Error('未配置硅基 API Key');
const apiCfg = options.apiConfig || getEmbeddingApiConfig();
const key = getApiKey(apiCfg.key);
if (!key) throw new Error('未配置 Embedding API Key');
const { timeout = 30000, signal } = options;
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), timeout);
try {
const response = await fetch(`${BASE_URL}/v1/embeddings`, {
const baseUrl = String(apiCfg.url || `${BASE_URL}/v1`).replace(/\/+$/, '');
const response = await fetch(`${baseUrl}/embeddings`, {
method: 'POST',
headers: {
'Authorization': `Bearer ${key}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
model: EMBEDDING_MODEL,
model: String(apiCfg.model || EMBEDDING_MODEL),
input: texts,
}),
signal: signal || controller.signal,