feat: maxToken setting

client/src/constants/model.ts

@@ -83,6 +83,7 @@ export const defaultModel: ModelSchema = {
     searchEmptyText: '',
     systemPrompt: '',
     temperature: 0,
+    maxToken: 4000,
     chatModel: OpenAiChatEnum.GPT35
   },
   share: {

client/src/pages/api/openapi/v1/chat/completions.ts

@@ -180,6 +180,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       await modelServiceToolMap[model.chat.chatModel].chatCompletion({
         apiKey: userOpenAiKey || apiKey,
         temperature: +temperature,
+        maxToken: model.chat.maxToken,
         messages: completePrompts,
         stream,
         res

client/src/pages/model/components/detail/components/Settings.tsx

@@ -36,11 +36,6 @@ const Settings = ({ modelId }: { modelId: string }) => {
   const [btnLoading, setBtnLoading] = useState(false);
   const [refresh, setRefresh] = useState(false);
 
-  const isOwner = useMemo(
-    () => modelDetail.userId === userInfo?._id,
-    [modelDetail.userId, userInfo?._id]
-  );
-
   const {
     register,
     setValue,
@@ -52,6 +47,20 @@ const Settings = ({ modelId }: { modelId: string }) => {
     defaultValues: modelDetail
   });
 
+  const isOwner = useMemo(
+    () => modelDetail.userId === userInfo?._id,
+    [modelDetail.userId, userInfo?._id]
+  );
+  const tokenLimit = useMemo(() => {
+    const max = ChatModelMap[getValues('chat.chatModel')]?.contextMaxToken || 4000;
+
+    if (max < getValues('chat.maxToken')) {
+      setValue('chat.maxToken', max);
+    }
+
+    return max;
+  }, [getValues, setValue, refresh]);
+
   // 提交保存模型修改
   const saveSubmitSuccess = useCallback(
     async (data: ModelSchema) => {
@@ -256,6 +265,27 @@ const Settings = ({ modelId }: { modelId: string }) => {
           />
         </Box>
       </Flex>
+      <Flex alignItems={'center'} mt={12} mb={10}>
+        <Box w={['60px', '100px', '140px']} flexShrink={0}>
+          最大长度
+        </Box>
+        <Box flex={1} ml={'10px'}>
+          <MySlider
+            markList={[
+              { label: '100', value: 100 },
+              { label: `${tokenLimit}`, value: tokenLimit }
+            ]}
+            width={['100%', '260px']}
+            min={100}
+            max={tokenLimit}
+            activeVal={getValues('chat.maxToken')}
+            setVal={(val) => {
+              setValue('chat.maxToken', val);
+              setRefresh(!refresh);
+            }}
+          />
+        </Box>
+      </Flex>
       <Flex mt={10} alignItems={'flex-start'}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
           提示词

client/src/service/models/model.ts

@@ -47,6 +47,11 @@ const ModelSchema = new Schema({
       type: String,
       default: ''
     },
+    maxToken: {
+      type: Number,
+      default: 4000,
+      min: 100
+    },
     temperature: {
       type: Number,
       min: 0,

client/src/service/utils/chat/index.ts

@@ -12,6 +12,7 @@ import { textAdaptGptResponse } from '@/utils/adapt';
 export type ChatCompletionType = {
   apiKey: string;
   temperature: number;
+  maxToken?: number;
   messages: ChatItemType[];
   chatId?: string;
   [key: string]: any;

client/src/service/utils/chat/openai.ts

@@ -19,22 +19,31 @@ export const chatResponse = async ({
   model,
   apiKey,
   temperature,
+  maxToken = 4000,
   messages,
   stream
 }: ChatCompletionType & { model: `${OpenAiChatEnum}` }) => {
+  const modelTokenLimit = ChatModelMap[model]?.contextMaxToken || 4000;
   const filterMessages = ChatContextFilter({
     model,
     prompts: messages,
-    maxTokens: Math.ceil(ChatModelMap[model].contextMaxToken * 0.85)
+    maxTokens: Math.ceil(modelTokenLimit - 300) // filter token. not response maxToken
   });
 
   const adaptMessages = adaptChatItem_openAI({ messages: filterMessages, reserveId: false });
   const chatAPI = getOpenAIApi();
 
+  const promptsToken = modelToolMap[model].countTokens({
+    messages: filterMessages
+  });
+
+  maxToken = maxToken + promptsToken > modelTokenLimit ? modelTokenLimit - promptsToken : maxToken;
+
   const response = await chatAPI.createChatCompletion(
     {
       model,
       temperature: Number(temperature || 0),
+      max_tokens: maxToken,
       messages: adaptMessages,
       frequency_penalty: 0.5, // 越大，重复内容越少
       presence_penalty: -0.5, // 越大，越容易出现新内容

client/src/types/mongoSchema.d.ts

@@ -44,6 +44,7 @@ export interface ModelSchema {
     searchEmptyText: string;
     systemPrompt: string;
     temperature: number;
+    maxToken: number;
     chatModel: ChatModelType; // 聊天时用的模型，训练后就是训练的模型
   };
   share: {