197 lines
6.7 KiB
JavaScript
197 lines
6.7 KiB
JavaScript
// ═══════════════════════════════════════════════════════════════════════════
|
||
// Reranker - 硅基 bge-reranker-v2-m3
|
||
// 对候选文档进行精排,过滤与 query 不相关的内容
|
||
// ═══════════════════════════════════════════════════════════════════════════
|
||
|
||
import { xbLog } from '../../../../core/debug-core.js';
|
||
import { getApiKey } from './siliconflow.js';
|
||
|
||
const MODULE_ID = 'reranker';
|
||
const RERANK_URL = 'https://api.siliconflow.cn/v1/rerank';
|
||
const RERANK_MODEL = 'BAAI/bge-reranker-v2-m3';
|
||
const DEFAULT_TIMEOUT = 15000;
|
||
const MAX_DOCUMENTS = 100; // API 限制
|
||
|
||
/**
|
||
* 对文档列表进行 Rerank 精排
|
||
*
|
||
* @param {string} query - 查询文本
|
||
* @param {Array<string>} documents - 文档文本列表
|
||
* @param {object} options - 选项
|
||
* @param {number} options.topN - 返回前 N 个结果,默认 40
|
||
* @param {number} options.timeout - 超时时间,默认 15000ms
|
||
* @param {AbortSignal} options.signal - 取消信号
|
||
* @returns {Promise<Array<{index: number, relevance_score: number}>>} 排序后的结果
|
||
*/
|
||
export async function rerank(query, documents, options = {}) {
|
||
const { topN = 40, timeout = DEFAULT_TIMEOUT, signal } = options;
|
||
|
||
if (!query?.trim()) {
|
||
xbLog.warn(MODULE_ID, 'query 为空,跳过 rerank');
|
||
return { results: documents.map((_, i) => ({ index: i, relevance_score: 0 })), failed: true };
|
||
}
|
||
|
||
if (!documents?.length) {
|
||
return { results: [], failed: false };
|
||
}
|
||
|
||
const key = getApiKey();
|
||
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 };
|
||
}
|
||
|
||
// 截断超长文档列表
|
||
const truncatedDocs = documents.slice(0, MAX_DOCUMENTS);
|
||
if (documents.length > MAX_DOCUMENTS) {
|
||
xbLog.warn(MODULE_ID, `文档数 ${documents.length} 超过限制 ${MAX_DOCUMENTS},已截断`);
|
||
}
|
||
|
||
// 过滤空文档,记录原始索引
|
||
const validDocs = [];
|
||
const indexMap = []; // validDocs index → original index
|
||
|
||
for (let i = 0; i < truncatedDocs.length; i++) {
|
||
const text = String(truncatedDocs[i] || '').trim();
|
||
if (text) {
|
||
validDocs.push(text);
|
||
indexMap.push(i);
|
||
}
|
||
}
|
||
|
||
if (!validDocs.length) {
|
||
xbLog.warn(MODULE_ID, '无有效文档,跳过 rerank');
|
||
return { results: [], failed: false };
|
||
}
|
||
|
||
const controller = new AbortController();
|
||
const timeoutId = setTimeout(() => controller.abort(), timeout);
|
||
|
||
try {
|
||
const T0 = performance.now();
|
||
|
||
const response = await fetch(RERANK_URL, {
|
||
method: 'POST',
|
||
headers: {
|
||
'Authorization': `Bearer ${key}`,
|
||
'Content-Type': 'application/json',
|
||
},
|
||
body: JSON.stringify({
|
||
model: RERANK_MODEL,
|
||
// Zero-darkbox: do not silently truncate query.
|
||
query,
|
||
documents: validDocs,
|
||
top_n: Math.min(topN, validDocs.length),
|
||
return_documents: false,
|
||
}),
|
||
signal: signal || controller.signal,
|
||
});
|
||
|
||
clearTimeout(timeoutId);
|
||
|
||
if (!response.ok) {
|
||
const errorText = await response.text().catch(() => '');
|
||
throw new Error(`Rerank API ${response.status}: ${errorText.slice(0, 200)}`);
|
||
}
|
||
|
||
const data = await response.json();
|
||
const results = data.results || [];
|
||
|
||
// 映射回原始索引
|
||
const mapped = results.map(r => ({
|
||
index: indexMap[r.index],
|
||
relevance_score: r.relevance_score ?? 0,
|
||
}));
|
||
|
||
const elapsed = Math.round(performance.now() - T0);
|
||
xbLog.info(MODULE_ID, `Rerank 完成: ${validDocs.length} docs → ${results.length} selected (${elapsed}ms)`);
|
||
|
||
return { results: mapped, failed: false };
|
||
|
||
} catch (e) {
|
||
clearTimeout(timeoutId);
|
||
|
||
if (e?.name === 'AbortError') {
|
||
xbLog.warn(MODULE_ID, 'Rerank 超时或取消');
|
||
} else {
|
||
xbLog.error(MODULE_ID, 'Rerank 失败', e);
|
||
}
|
||
|
||
// 降级:返回原顺序,分数均匀分布
|
||
return {
|
||
results: documents.slice(0, topN).map((_, i) => ({
|
||
index: i,
|
||
relevance_score: 0,
|
||
})),
|
||
failed: true,
|
||
};
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 对 chunk 对象列表进行 Rerank
|
||
*
|
||
* @param {string} query - 查询文本
|
||
* @param {Array<object>} chunks - chunk 对象列表,需要有 text 字段
|
||
* @param {object} options - 选项
|
||
* @returns {Promise<Array<object>>} 排序后的 chunk 列表,带 _rerankScore 字段
|
||
*/
|
||
export async function rerankChunks(query, chunks, options = {}) {
|
||
const { topN = 40, minScore = 0.1 } = options;
|
||
|
||
if (!chunks?.length) return [];
|
||
if (chunks.length <= topN) {
|
||
const texts = chunks.map(c => c.text || c.semantic || '');
|
||
const { results, failed } = await rerank(query, texts, { topN: chunks.length, ...options });
|
||
|
||
if (failed) {
|
||
return chunks.map(c => ({ ...c, _rerankScore: 0, _rerankFailed: true }));
|
||
}
|
||
|
||
const scoreMap = new Map(results.map(r => [r.index, r.relevance_score]));
|
||
return chunks.map((c, i) => ({
|
||
...c,
|
||
_rerankScore: scoreMap.get(i) ?? 0,
|
||
})).sort((a, b) => b._rerankScore - a._rerankScore);
|
||
}
|
||
|
||
const texts = chunks.map(c => c.text || c.semantic || '');
|
||
const { results, failed } = await rerank(query, texts, { topN, ...options });
|
||
|
||
if (failed) {
|
||
return chunks.slice(0, topN).map(c => ({
|
||
...c,
|
||
_rerankScore: 0,
|
||
_rerankFailed: true,
|
||
}));
|
||
}
|
||
|
||
return results
|
||
.filter(r => r.relevance_score >= minScore)
|
||
.sort((a, b) => b.relevance_score - a.relevance_score)
|
||
.map(r => ({
|
||
...chunks[r.index],
|
||
_rerankScore: r.relevance_score,
|
||
}));
|
||
}
|
||
|
||
/**
|
||
* 测试 Rerank 服务连接
|
||
*/
|
||
export async function testRerankService() {
|
||
const key = getApiKey();
|
||
if (!key) {
|
||
throw new Error('请配置硅基 API Key');
|
||
}
|
||
|
||
try {
|
||
const { results } = await rerank('测试查询', ['测试文档1', '测试文档2'], { topN: 2 });
|
||
return {
|
||
success: true,
|
||
message: `连接成功,返回 ${results.length} 个结果`,
|
||
};
|
||
} catch (e) {
|
||
throw new Error(`连接失败: ${e.message}`);
|
||
}
|
||
}
|