mirror of
https://github.com/labring/FastGPT.git
synced 2025-07-27 08:25:07 +00:00
perf: search kb model
This commit is contained in:
@@ -57,61 +57,28 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
||||
|
||||
// 使用了知识库搜索
|
||||
if (model.chat.useKb) {
|
||||
const { systemPrompts } = await searchKb_openai({
|
||||
const { code, searchPrompt } = await searchKb_openai({
|
||||
apiKey: userApiKey || systemKey,
|
||||
isPay: !userApiKey,
|
||||
text: prompt.value,
|
||||
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22,
|
||||
modelId,
|
||||
model,
|
||||
userId
|
||||
});
|
||||
|
||||
// filter system prompt
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
) {
|
||||
return res.send('对不起,你的问题不在知识库中。');
|
||||
// search result is empty
|
||||
if (code === 201) {
|
||||
return res.send(searchPrompt?.value);
|
||||
}
|
||||
/* 高相似度+无上下文,不添加额外知识,仅用系统提示词 */
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.noContext
|
||||
) {
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: model.chat.systemPrompt
|
||||
});
|
||||
} else {
|
||||
// 有匹配情况下,system 添加知识库内容。
|
||||
// 系统提示词过滤,最多 2500 tokens
|
||||
const filterSystemPrompt = systemPromptFilter({
|
||||
model: model.chat.chatModel,
|
||||
prompts: systemPrompts,
|
||||
maxTokens: 2500
|
||||
});
|
||||
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: `
|
||||
${model.chat.systemPrompt}
|
||||
${
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
? `不回答知识库外的内容.`
|
||||
: ''
|
||||
}
|
||||
知识库内容为: ${filterSystemPrompt}'
|
||||
`
|
||||
});
|
||||
}
|
||||
searchPrompt && prompts.unshift(searchPrompt);
|
||||
} else {
|
||||
// 没有用知识库搜索,仅用系统提示词
|
||||
if (model.chat.systemPrompt) {
|
||||
model.chat.systemPrompt &&
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: model.chat.systemPrompt
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
// 控制总 tokens 数量,防止超出
|
||||
|
@@ -67,57 +67,21 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
||||
if (model.chat.useKb) {
|
||||
const similarity = ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22;
|
||||
|
||||
const { systemPrompts } = await searchKb_openai({
|
||||
const { code, searchPrompt } = await searchKb_openai({
|
||||
apiKey,
|
||||
isPay: true,
|
||||
text: prompts[prompts.length - 1].value,
|
||||
similarity,
|
||||
modelId,
|
||||
model,
|
||||
userId
|
||||
});
|
||||
|
||||
// filter system prompt
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
) {
|
||||
return jsonRes(res, {
|
||||
code: 500,
|
||||
message: '对不起,你的问题不在知识库中。',
|
||||
data: '对不起,你的问题不在知识库中。'
|
||||
});
|
||||
// search result is empty
|
||||
if (code === 201) {
|
||||
return res.send(searchPrompt?.value);
|
||||
}
|
||||
/* 高相似度+无上下文,不添加额外知识,仅用系统提示词 */
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.noContext
|
||||
) {
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: model.chat.systemPrompt
|
||||
});
|
||||
} else {
|
||||
// 有匹配情况下,system 添加知识库内容。
|
||||
// 系统提示词过滤,最多 2500 tokens
|
||||
const filterSystemPrompt = systemPromptFilter({
|
||||
model: model.chat.chatModel,
|
||||
prompts: systemPrompts,
|
||||
maxTokens: 2500
|
||||
});
|
||||
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: `
|
||||
${model.chat.systemPrompt}
|
||||
${
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
? `不回答知识库外的内容.`
|
||||
: ''
|
||||
}
|
||||
知识库内容为: ${filterSystemPrompt}'
|
||||
`
|
||||
});
|
||||
}
|
||||
searchPrompt && prompts.unshift(searchPrompt);
|
||||
} else {
|
||||
// 没有用知识库搜索,仅用系统提示词
|
||||
if (model.chat.systemPrompt) {
|
||||
|
@@ -131,26 +131,16 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
||||
const prompts = [prompt];
|
||||
|
||||
// 获取向量匹配到的提示词
|
||||
const { systemPrompts } = await searchKb_openai({
|
||||
const { searchPrompt } = await searchKb_openai({
|
||||
isPay: true,
|
||||
apiKey,
|
||||
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22,
|
||||
text: prompt.value,
|
||||
modelId,
|
||||
model,
|
||||
userId
|
||||
});
|
||||
|
||||
// system 筛选,最多 2500 tokens
|
||||
const filterSystemPrompt = systemPromptFilter({
|
||||
model: model.chat.chatModel,
|
||||
prompts: systemPrompts,
|
||||
maxTokens: 2500
|
||||
});
|
||||
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: `${model.chat.systemPrompt} 知识库是最新的,下面是知识库内容:${filterSystemPrompt}`
|
||||
});
|
||||
searchPrompt && prompts.unshift(searchPrompt);
|
||||
|
||||
// 控制上下文 tokens 数量,防止超出
|
||||
const filterPrompts = openaiChatFilter({
|
||||
@@ -181,8 +171,6 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
||||
}
|
||||
);
|
||||
|
||||
console.log('code response. time:', `${(Date.now() - startTime) / 1000}s`);
|
||||
|
||||
let responseContent = '';
|
||||
|
||||
if (isStream) {
|
||||
|
@@ -68,60 +68,21 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
||||
}
|
||||
|
||||
// 获取向量匹配到的提示词
|
||||
const { systemPrompts } = await searchKb_openai({
|
||||
const { code, searchPrompt } = await searchKb_openai({
|
||||
isPay: true,
|
||||
apiKey,
|
||||
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22,
|
||||
text: prompts[prompts.length - 1].value,
|
||||
modelId,
|
||||
model,
|
||||
userId
|
||||
});
|
||||
|
||||
// system 合并
|
||||
if (prompts[0].obj === 'SYSTEM') {
|
||||
systemPrompts.unshift(prompts.shift()?.value || '');
|
||||
// search result is empty
|
||||
if (code === 201) {
|
||||
return res.send(searchPrompt?.value);
|
||||
}
|
||||
|
||||
/* 高相似度+退出,无法匹配时直接退出 */
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
) {
|
||||
return jsonRes(res, {
|
||||
code: 500,
|
||||
message: '对不起,你的问题不在知识库中。',
|
||||
data: '对不起,你的问题不在知识库中。'
|
||||
});
|
||||
}
|
||||
/* 高相似度+无上下文,不添加额外知识 */
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.noContext
|
||||
) {
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: model.chat.systemPrompt
|
||||
});
|
||||
} else {
|
||||
// 有匹配或者低匹配度模式情况下,添加知识库内容。
|
||||
// 系统提示词过滤,最多 2500 tokens
|
||||
const systemPrompt = systemPromptFilter({
|
||||
model: model.chat.chatModel,
|
||||
prompts: systemPrompts,
|
||||
maxTokens: 2500
|
||||
});
|
||||
|
||||
prompts.unshift({
|
||||
obj: 'SYSTEM',
|
||||
value: `
|
||||
${model.chat.systemPrompt}
|
||||
${
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity ? `不回答知识库外的内容.` : ''
|
||||
}
|
||||
知识库内容为: ${systemPrompt}'
|
||||
`
|
||||
});
|
||||
}
|
||||
searchPrompt && prompts.unshift(searchPrompt);
|
||||
|
||||
// 控制在 tokens 数量,防止超出
|
||||
const filterPrompts = openaiChatFilter({
|
||||
|
@@ -1,6 +1,8 @@
|
||||
import { openaiCreateEmbedding } from '../utils/openai';
|
||||
import { PgClient } from '@/service/pg';
|
||||
import { ModelDataStatusEnum } from '@/constants/model';
|
||||
import { ModelDataStatusEnum, ModelVectorSearchModeEnum } from '@/constants/model';
|
||||
import { ModelSchema } from '@/types/mongoSchema';
|
||||
import { systemPromptFilter } from '../utils/tools';
|
||||
|
||||
/**
|
||||
* use openai embedding search kb
|
||||
@@ -10,16 +12,22 @@ export const searchKb_openai = async ({
|
||||
isPay,
|
||||
text,
|
||||
similarity,
|
||||
modelId,
|
||||
model,
|
||||
userId
|
||||
}: {
|
||||
apiKey: string;
|
||||
isPay: boolean;
|
||||
text: string;
|
||||
modelId: string;
|
||||
model: ModelSchema;
|
||||
userId: string;
|
||||
similarity: number;
|
||||
}) => {
|
||||
}): Promise<{
|
||||
code: 200 | 201;
|
||||
searchPrompt?: {
|
||||
obj: 'Human' | 'AI' | 'SYSTEM';
|
||||
value: string;
|
||||
};
|
||||
}> => {
|
||||
// 获取提示词的向量
|
||||
const { vector: promptVector } = await openaiCreateEmbedding({
|
||||
isPay,
|
||||
@@ -28,12 +36,12 @@ export const searchKb_openai = async ({
|
||||
text
|
||||
});
|
||||
|
||||
const vectorSearch = await PgClient.select<{ id: string; q: string; a: string }>('modelData', {
|
||||
fields: ['id', 'q', 'a'],
|
||||
const vectorSearch = await PgClient.select<{ q: string; a: string }>('modelData', {
|
||||
fields: ['q', 'a'],
|
||||
where: [
|
||||
['status', ModelDataStatusEnum.ready],
|
||||
'AND',
|
||||
['model_id', modelId],
|
||||
['model_id', model._id],
|
||||
'AND',
|
||||
`vector <=> '[${promptVector}]' < ${similarity}`
|
||||
],
|
||||
@@ -43,5 +51,51 @@ export const searchKb_openai = async ({
|
||||
|
||||
const systemPrompts: string[] = vectorSearch.rows.map((item) => `${item.q}\n${item.a}`);
|
||||
|
||||
return { systemPrompts };
|
||||
// filter system prompt
|
||||
if (
|
||||
systemPrompts.length === 0 &&
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
|
||||
) {
|
||||
return {
|
||||
code: 201,
|
||||
searchPrompt: {
|
||||
obj: 'AI',
|
||||
value: '对不起,你的问题不在知识库中。'
|
||||
}
|
||||
};
|
||||
}
|
||||
/* 高相似度+无上下文,不添加额外知识,仅用系统提示词 */
|
||||
if (systemPrompts.length === 0 && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
|
||||
return {
|
||||
code: 200,
|
||||
searchPrompt: model.chat.systemPrompt
|
||||
? {
|
||||
obj: 'SYSTEM',
|
||||
value: model.chat.systemPrompt
|
||||
}
|
||||
: undefined
|
||||
};
|
||||
}
|
||||
|
||||
// 有匹配情况下,system 添加知识库内容。
|
||||
// 系统提示词过滤,最多 2500 tokens
|
||||
const filterSystemPrompt = systemPromptFilter({
|
||||
model: model.chat.chatModel,
|
||||
prompts: systemPrompts,
|
||||
maxTokens: 2500
|
||||
});
|
||||
|
||||
return {
|
||||
code: 200,
|
||||
searchPrompt: {
|
||||
obj: 'SYSTEM',
|
||||
value: `
|
||||
${model.chat.systemPrompt}
|
||||
${
|
||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity ? `不回答知识库外的内容.` : ''
|
||||
}
|
||||
知识库内容为: ${filterSystemPrompt}'
|
||||
`
|
||||
}
|
||||
};
|
||||
};
|
||||
|
Reference in New Issue
Block a user