opensource/sqlchat
1
0
Fork 0
mirror of https://github.com/sqlchat/sqlchat.git synced 2025-09-29 02:54:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: set different token limit per model

Browse Source
This commit is contained in:
tianzhou
2023-05-31 16:47:13 +08:00
parent 933fb6f280
commit 088f0904c0
2 changed files with 6 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/components/ConversationView/index.tsx
View File

@ -14,7 +14,7 @@ import {
useUserStore, useUserStore,
} from "@/store"; } from "@/store";
import { Conversation, CreatorRole, Message } from "@/types"; import { Conversation, CreatorRole, Message } from "@/types";
import { countTextTokens, generateUUID, hasFeature } from "@/utils"; import { countTextTokens, generateUUID, getModel, hasFeature } from "@/utils";
import getEventEmitter from "@/utils/event-emitter"; import getEventEmitter from "@/utils/event-emitter";
import Header from "./Header"; import Header from "./Header";
import EmptyView from "../EmptyView"; import EmptyView from "../EmptyView";
@ -23,10 +23,6 @@ import ClearConversationButton from "../ClearConversationButton";
import MessageTextarea from "./MessageTextarea"; import MessageTextarea from "./MessageTextarea";
import DataStorageBanner from "../DataStorageBanner"; import DataStorageBanner from "../DataStorageBanner";
// The maximum number of tokens that can be sent to the OpenAI API.
// reference: https://platform.openai.com/docs/api-reference/completions/create#completions/create-max_tokens
const MAX_TOKENS = 4000;
const ConversationView = () => { const ConversationView = () => {
const { data: session } = useSession(); const { data: session } = useSession();
const settingStore = useSettingStore(); const settingStore = useSettingStore();
@ -135,6 +131,7 @@ const ConversationView = () => {
const messageList = messageStore.getState().messageList.filter((message: Message) => message.conversationId === currentConversation.id); const messageList = messageStore.getState().messageList.filter((message: Message) => message.conversationId === currentConversation.id);
const promptGenerator = getPromptGeneratorOfAssistant(getAssistantById(currentConversation.assistantId)!); const promptGenerator = getPromptGeneratorOfAssistant(getAssistantById(currentConversation.assistantId)!);
let dbPrompt = promptGenerator(); let dbPrompt = promptGenerator();
const maxToken = getModel(settingStore.setting.openAIApiConfig?.model || "").max_token;
// Squeeze as much prompt as possible under the token limit, the prompt is in the order of: // Squeeze as much prompt as possible under the token limit, the prompt is in the order of:
// 1. Assistant specific prompt with database schema if applicable. // 1. Assistant specific prompt with database schema if applicable.
// 2. A list of previous exchanges. // 2. A list of previous exchanges.
@ -166,7 +163,7 @@ const ConversationView = () => {
} }
if (tableList) { if (tableList) {
for (const table of tableList) { for (const table of tableList) {
if (tokens < MAX_TOKENS / 2) { if (tokens < maxToken / 2) {
tokens += countTextTokens(table); tokens += countTextTokens(table);
schema += table; schema += table;
} }
@ -185,7 +182,7 @@ const ConversationView = () => {
for (let i = messageList.length - 1; i >= 0; i--) { for (let i = messageList.length - 1; i >= 0; i--) {
const message = messageList[i]; const message = messageList[i];
if (message.status === "DONE") { if (message.status === "DONE") {
if (tokens < MAX_TOKENS) { if (tokens < maxToken) {
tokens += countTextTokens(message.content); tokens += countTextTokens(message.content);
formatedMessageList.unshift({ formatedMessageList.unshift({
role: message.creatorRole, role: message.creatorRole,

2
src/utils/model.ts
View File

@ -3,6 +3,7 @@ const gpt35 = {
temperature: 0, temperature: 0,
frequency_penalty: 0.0, frequency_penalty: 0.0,
presence_penalty: 0.0, presence_penalty: 0.0,
max_token: 4000,
}; };
const gpt4 = { const gpt4 = {
@ -10,6 +11,7 @@ const gpt4 = {
temperature: 0, temperature: 0,
frequency_penalty: 0.0, frequency_penalty: 0.0,
presence_penalty: 0.0, presence_penalty: 0.0,
max_token: 8000,
}; };
export const models = [gpt35, gpt4]; export const models = [gpt35, gpt4];