perf: search prompt
This commit is contained in:
parent
f52f514f5f
commit
4dc541e0a6
@ -35,7 +35,7 @@ const Layout = ({ children, isPcDevice }: { children: JSX.Element; isPcDevice: b
|
|||||||
|
|
||||||
return (
|
return (
|
||||||
<>
|
<>
|
||||||
<Box h={'100%'} overflow={'overlay'} bg={'gray.100'}>
|
<Box h={'100%'} overflowY={'auto'} bg={'gray.100'}>
|
||||||
{isPc ? (
|
{isPc ? (
|
||||||
pcUnShowLayoutRoute[router.pathname] ? (
|
pcUnShowLayoutRoute[router.pathname] ? (
|
||||||
<Auth>{children}</Auth>
|
<Auth>{children}</Auth>
|
||||||
|
|||||||
@ -102,9 +102,7 @@ const Navbar = () => {
|
|||||||
justifyContent={'center'}
|
justifyContent={'center'}
|
||||||
onClick={() => {
|
onClick={() => {
|
||||||
if (item.link === router.asPath) return;
|
if (item.link === router.asPath) return;
|
||||||
router.push(item.link, undefined, {
|
router.push(item.link);
|
||||||
shallow: true
|
|
||||||
});
|
|
||||||
}}
|
}}
|
||||||
cursor={'pointer'}
|
cursor={'pointer'}
|
||||||
w={'60px'}
|
w={'60px'}
|
||||||
|
|||||||
@ -55,7 +55,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
|
|
||||||
// 使用了知识库搜索
|
// 使用了知识库搜索
|
||||||
if (model.chat.useKb) {
|
if (model.chat.useKb) {
|
||||||
const { code, searchPrompt } = await searchKb({
|
const { code, searchPrompts } = await searchKb({
|
||||||
userOpenAiKey,
|
userOpenAiKey,
|
||||||
prompts,
|
prompts,
|
||||||
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
|
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
|
||||||
@ -65,10 +65,10 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
|
|
||||||
// search result is empty
|
// search result is empty
|
||||||
if (code === 201) {
|
if (code === 201) {
|
||||||
return res.send(searchPrompt?.value);
|
return res.send(searchPrompts[0]?.value);
|
||||||
}
|
}
|
||||||
|
|
||||||
searchPrompt && prompts.unshift(searchPrompt);
|
prompts.splice(prompts.length - 1, 0, ...searchPrompts);
|
||||||
} else {
|
} else {
|
||||||
// 没有用知识库搜索,仅用系统提示词
|
// 没有用知识库搜索,仅用系统提示词
|
||||||
model.chat.systemPrompt &&
|
model.chat.systemPrompt &&
|
||||||
@ -103,8 +103,12 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
stream,
|
stream,
|
||||||
chatResponse: streamResponse,
|
chatResponse: streamResponse,
|
||||||
prompts,
|
prompts,
|
||||||
systemPrompt:
|
systemPrompt: showModelDetail
|
||||||
showModelDetail && prompts[0].obj === ChatRoleEnum.System ? prompts[0].value : ''
|
? prompts
|
||||||
|
.filter((item) => item.obj === ChatRoleEnum.System)
|
||||||
|
.map((item) => item.value)
|
||||||
|
.join('\n')
|
||||||
|
: ''
|
||||||
});
|
});
|
||||||
|
|
||||||
// 只有使用平台的 key 才计费
|
// 只有使用平台的 key 才计费
|
||||||
|
|||||||
@ -73,7 +73,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
if (model.chat.useKb) {
|
if (model.chat.useKb) {
|
||||||
const similarity = ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22;
|
const similarity = ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22;
|
||||||
|
|
||||||
const { code, searchPrompt } = await searchKb({
|
const { code, searchPrompts } = await searchKb({
|
||||||
prompts,
|
prompts,
|
||||||
similarity,
|
similarity,
|
||||||
model,
|
model,
|
||||||
@ -82,10 +82,9 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
|
|
||||||
// search result is empty
|
// search result is empty
|
||||||
if (code === 201) {
|
if (code === 201) {
|
||||||
return res.send(searchPrompt?.value);
|
return res.send(searchPrompts[0]?.value);
|
||||||
}
|
}
|
||||||
|
prompts.splice(prompts.length - 1, 0, ...searchPrompts);
|
||||||
searchPrompt && prompts.unshift(searchPrompt);
|
|
||||||
} else {
|
} else {
|
||||||
// 没有用知识库搜索,仅用系统提示词
|
// 没有用知识库搜索,仅用系统提示词
|
||||||
if (model.chat.systemPrompt) {
|
if (model.chat.systemPrompt) {
|
||||||
|
|||||||
@ -122,14 +122,14 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
|
|||||||
const prompts = [prompt];
|
const prompts = [prompt];
|
||||||
|
|
||||||
// 获取向量匹配到的提示词
|
// 获取向量匹配到的提示词
|
||||||
const { searchPrompt } = await searchKb({
|
const { searchPrompts } = await searchKb({
|
||||||
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
|
similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
|
||||||
prompts,
|
prompts,
|
||||||
model,
|
model,
|
||||||
userId
|
userId
|
||||||
});
|
});
|
||||||
|
|
||||||
searchPrompt && prompts.unshift(searchPrompt);
|
prompts.splice(prompts.length - 1, 0, ...searchPrompts);
|
||||||
|
|
||||||
// 计算温度
|
// 计算温度
|
||||||
const temperature = (modelConstantsData.maxTemperature * (model.chat.temperature / 10)).toFixed(
|
const temperature = (modelConstantsData.maxTemperature * (model.chat.temperature / 10)).toFixed(
|
||||||
|
|||||||
@ -508,7 +508,7 @@ const Chat = ({
|
|||||||
isLeavePage.current = true;
|
isLeavePage.current = true;
|
||||||
controller.current?.abort();
|
controller.current?.abort();
|
||||||
};
|
};
|
||||||
}, []);
|
}, [modelId, chatId]);
|
||||||
|
|
||||||
return (
|
return (
|
||||||
<Flex
|
<Flex
|
||||||
|
|||||||
@ -279,9 +279,9 @@ const ModelDataCard = ({ modelId, isOwner }: { modelId: string; isOwner: boolean
|
|||||||
</Tbody>
|
</Tbody>
|
||||||
</Table>
|
</Table>
|
||||||
</TableContainer>
|
</TableContainer>
|
||||||
<Box mt={2} textAlign={'end'}>
|
<Flex mt={2} justifyContent={'flex-end'}>
|
||||||
<Pagination />
|
<Pagination />
|
||||||
</Box>
|
</Flex>
|
||||||
</Box>
|
</Box>
|
||||||
|
|
||||||
<Loading loading={isLoading} fixed={false} />
|
<Loading loading={isLoading} fixed={false} />
|
||||||
|
|||||||
@ -61,13 +61,7 @@ const SelectFileModal = ({
|
|||||||
const { openConfirm, ConfirmChild } = useConfirm({
|
const { openConfirm, ConfirmChild } = useConfirm({
|
||||||
content: `确认导入该文件,需要一定时间进行拆解,该任务无法终止!如果余额不足,未完成的任务会被直接清除。一共 ${
|
content: `确认导入该文件,需要一定时间进行拆解,该任务无法终止!如果余额不足,未完成的任务会被直接清除。一共 ${
|
||||||
splitRes.chunks.length
|
splitRes.chunks.length
|
||||||
} 组。${
|
} 组。${splitRes.tokens ? `大约 ${splitRes.tokens} 个tokens。` : ''}`
|
||||||
splitRes.tokens
|
|
||||||
? `大约 ${splitRes.tokens} 个tokens, 约 ${formatPrice(
|
|
||||||
splitRes.tokens * modeMap[mode].price
|
|
||||||
)} 元`
|
|
||||||
: ''
|
|
||||||
}`
|
|
||||||
});
|
});
|
||||||
|
|
||||||
const onSelectFile = useCallback(
|
const onSelectFile = useCallback(
|
||||||
|
|||||||
@ -101,9 +101,9 @@ const modelList = () => {
|
|||||||
<Grid templateColumns={['1fr', '1fr 1fr', '1fr 1fr 1fr']} gridGap={4} mt={4}>
|
<Grid templateColumns={['1fr', '1fr 1fr', '1fr 1fr 1fr']} gridGap={4} mt={4}>
|
||||||
<ShareModelList models={models} onclickCollection={onclickCollection} />
|
<ShareModelList models={models} onclickCollection={onclickCollection} />
|
||||||
</Grid>
|
</Grid>
|
||||||
<Box mt={4}>
|
<Flex mt={4} justifyContent={'flex-end'}>
|
||||||
<Pagination />
|
<Pagination />
|
||||||
</Box>
|
</Flex>
|
||||||
</Card>
|
</Card>
|
||||||
|
|
||||||
<Loading loading={isLoading} />
|
<Loading loading={isLoading} />
|
||||||
|
|||||||
@ -1,5 +1,5 @@
|
|||||||
import React from 'react';
|
import React from 'react';
|
||||||
import { Card, Box, Table, Thead, Tbody, Tr, Th, Td, TableContainer } from '@chakra-ui/react';
|
import { Card, Box, Table, Thead, Tbody, Tr, Th, Td, TableContainer, Flex } from '@chakra-ui/react';
|
||||||
import { BillTypeMap } from '@/constants/user';
|
import { BillTypeMap } from '@/constants/user';
|
||||||
import { getUserBills } from '@/api/user';
|
import { getUserBills } from '@/api/user';
|
||||||
import type { UserBillType } from '@/types/user';
|
import type { UserBillType } from '@/types/user';
|
||||||
@ -48,9 +48,9 @@ const BillTable = () => {
|
|||||||
|
|
||||||
<Loading loading={isLoading} fixed={false} />
|
<Loading loading={isLoading} fixed={false} />
|
||||||
</TableContainer>
|
</TableContainer>
|
||||||
<Box mt={4} mr={4} textAlign={'end'}>
|
<Flex mt={4} px={4} justifyContent={'flex-end'}>
|
||||||
<Pagination />
|
<Pagination />
|
||||||
</Box>
|
</Flex>
|
||||||
</Card>
|
</Card>
|
||||||
);
|
);
|
||||||
};
|
};
|
||||||
|
|||||||
@ -139,9 +139,9 @@ const OpenApi = () => {
|
|||||||
|
|
||||||
<Loading loading={isLoading} fixed={false} />
|
<Loading loading={isLoading} fixed={false} />
|
||||||
</TableContainer>
|
</TableContainer>
|
||||||
<Box mt={4} mr={4} textAlign={'end'}>
|
<Flex mt={4} px={4} justifyContent={'flex-end'}>
|
||||||
<Pagination />
|
<Pagination />
|
||||||
</Box>
|
</Flex>
|
||||||
</Card>
|
</Card>
|
||||||
<Modal isOpen={isOpenWithdraw} onClose={onCloseWithdraw}>
|
<Modal isOpen={isOpenWithdraw} onClose={onCloseWithdraw}>
|
||||||
<ModalOverlay />
|
<ModalOverlay />
|
||||||
|
|||||||
@ -23,10 +23,10 @@ export const searchKb = async ({
|
|||||||
similarity?: number;
|
similarity?: number;
|
||||||
}): Promise<{
|
}): Promise<{
|
||||||
code: 200 | 201;
|
code: 200 | 201;
|
||||||
searchPrompt?: {
|
searchPrompts: {
|
||||||
obj: `${ChatRoleEnum}`;
|
obj: ChatRoleEnum;
|
||||||
value: string;
|
value: string;
|
||||||
};
|
}[];
|
||||||
}> => {
|
}> => {
|
||||||
async function search(textArr: string[] = []) {
|
async function search(textArr: string[] = []) {
|
||||||
// 获取提示词的向量
|
// 获取提示词的向量
|
||||||
@ -85,17 +85,38 @@ export const searchKb = async ({
|
|||||||
};
|
};
|
||||||
const filterRate = filterRateMap[systemPrompts.length] || filterRateMap[0];
|
const filterRate = filterRateMap[systemPrompts.length] || filterRateMap[0];
|
||||||
|
|
||||||
// count fixed system prompt
|
// 计算固定提示词的 token 数量
|
||||||
const fixedSystemPrompt = `
|
const fixedPrompts = [
|
||||||
${model.chat.systemPrompt}
|
...(model.chat.systemPrompt
|
||||||
${
|
? [
|
||||||
model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity ? '不回答知识库外的内容.' : ''
|
{
|
||||||
|
obj: ChatRoleEnum.System,
|
||||||
|
value: model.chat.systemPrompt
|
||||||
}
|
}
|
||||||
知识库内容为:`;
|
]
|
||||||
|
: []),
|
||||||
|
...(model.chat.searchMode === ModelVectorSearchModeEnum.noContext
|
||||||
|
? [
|
||||||
|
{
|
||||||
|
obj: ChatRoleEnum.System,
|
||||||
|
value: `知识库是关于"${model.name}"的内容,根据知识库内容回答问题.`
|
||||||
|
}
|
||||||
|
]
|
||||||
|
: [
|
||||||
|
{
|
||||||
|
obj: ChatRoleEnum.System,
|
||||||
|
value: `我们来玩问答游戏,规则为:
|
||||||
|
1.你忘记你已有的知识
|
||||||
|
2.你只能回答关于"${model.name}"的问题
|
||||||
|
3.你只能从知识库中选择内容进行回答
|
||||||
|
4.如果问题不在知识库中,你会回答"我不知道。"
|
||||||
|
务必遵守规则`
|
||||||
|
}
|
||||||
|
])
|
||||||
|
];
|
||||||
const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
|
const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
|
||||||
messages: [{ obj: 'System', value: fixedSystemPrompt }]
|
messages: fixedPrompts
|
||||||
});
|
});
|
||||||
|
|
||||||
const maxTokens = modelConstantsData.systemMaxToken - fixedSystemTokens;
|
const maxTokens = modelConstantsData.systemMaxToken - fixedSystemTokens;
|
||||||
|
|
||||||
const filterSystemPrompt = filterRate
|
const filterSystemPrompt = filterRate
|
||||||
@ -111,31 +132,38 @@ ${
|
|||||||
if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity) {
|
if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity) {
|
||||||
return {
|
return {
|
||||||
code: 201,
|
code: 201,
|
||||||
searchPrompt: {
|
searchPrompts: [
|
||||||
obj: ChatRoleEnum.AI,
|
{
|
||||||
|
obj: ChatRoleEnum.System,
|
||||||
value: '对不起,你的问题不在知识库中。'
|
value: '对不起,你的问题不在知识库中。'
|
||||||
}
|
}
|
||||||
|
]
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
/* 高相似度+无上下文,不添加额外知识,仅用系统提示词 */
|
/* 高相似度+无上下文,不添加额外知识,仅用系统提示词 */
|
||||||
if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
|
if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
|
||||||
return {
|
return {
|
||||||
code: 200,
|
code: 200,
|
||||||
searchPrompt: model.chat.systemPrompt
|
searchPrompts: model.chat.systemPrompt
|
||||||
? {
|
? [
|
||||||
|
{
|
||||||
obj: ChatRoleEnum.System,
|
obj: ChatRoleEnum.System,
|
||||||
value: model.chat.systemPrompt
|
value: model.chat.systemPrompt
|
||||||
}
|
}
|
||||||
: undefined
|
]
|
||||||
|
: []
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
/* 有匹配 */
|
/* 有匹配 */
|
||||||
return {
|
return {
|
||||||
code: 200,
|
code: 200,
|
||||||
searchPrompt: {
|
searchPrompts: [
|
||||||
|
{
|
||||||
obj: ChatRoleEnum.System,
|
obj: ChatRoleEnum.System,
|
||||||
value: `${fixedSystemPrompt}'${filterSystemPrompt}'`
|
value: `知识库:'${filterSystemPrompt}'`
|
||||||
}
|
},
|
||||||
|
...fixedPrompts
|
||||||
|
]
|
||||||
};
|
};
|
||||||
};
|
};
|
||||||
|
|||||||
@ -27,7 +27,7 @@ export const lafClaudChat = async ({
|
|||||||
.join('\n');
|
.join('\n');
|
||||||
const systemPromptText = systemPrompt ? `\n知识库内容:'${systemPrompt}'\n` : '';
|
const systemPromptText = systemPrompt ? `\n知识库内容:'${systemPrompt}'\n` : '';
|
||||||
|
|
||||||
const prompt = `${systemPromptText}我的问题:'${messages[messages.length - 1].value}'`;
|
const prompt = `${systemPromptText}\n我的问题是:'${messages[messages.length - 1].value}'`;
|
||||||
|
|
||||||
const lafResponse = await axios.post(
|
const lafResponse = await axios.post(
|
||||||
'https://hnvacz.laf.run/claude-gpt',
|
'https://hnvacz.laf.run/claude-gpt',
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user