add param max_gen_tokens

src/app.tsx

@@ -29,6 +29,8 @@ export interface ChatStore {
   tokenMargin: number;
   totalTokens: number;
   maxTokens: number;
+  maxGenTokens: number;
+  maxGenTokens_enabled: boolean;
   apiKey: string;
   apiEndpoint: string;
   streamMode: boolean;
@@ -75,6 +77,8 @@ export const newChatStore = (
     tokenMargin: 1024,
     totalTokens: 0,
     maxTokens: models[getDefaultParams("model", model)]?.maxToken ?? 4096,
+    maxGenTokens: 2048,
+    maxGenTokens_enabled: true,
     apiKey: getDefaultParams("key", apiKey),
     apiEndpoint: getDefaultParams("api", apiEndpoint),
     streamMode: getDefaultParams("mode", streamMode),
@@ -165,6 +169,8 @@ export function App() {
     const ret: ChatStore = await (await db).get(STORAGE_NAME, index);
     if (ret === null || ret === undefined) return newChatStore();
     // handle read from old version chatstore
+    if (ret.maxGenTokens === undefined) ret.maxGenTokens = 2048;
+    if (ret.maxGenTokens_enabled === undefined) ret.maxGenTokens_enabled = true;
     if (ret.model === undefined) ret.model = "gpt-3.5-turbo";
     if (ret.responseModelName === undefined) ret.responseModelName = "";
     if (ret.chatgpt_api_web_version === undefined)

src/chatbox.tsx

@@ -184,6 +184,8 @@ export default function ChatBOX(props: {
       });
     client.model = chatStore.model;
     client.max_tokens = chatStore.maxTokens;
+    client.max_gen_tokens = chatStore.maxGenTokens;
+    client.enable_max_gen_tokens = chatStore.maxGenTokens_enabled;
 
     try {
       setShowGenerating(true);

src/chatgpt.ts

@@ -63,6 +63,8 @@ class Chat {
   sysMessageContent: string;
   total_tokens: number;
   max_tokens: number;
+  max_gen_tokens: number;
+  enable_max_gen_tokens: boolean;
   tokens_margin: number;
   apiEndpoint: string;
   model: string;
@@ -78,6 +80,8 @@ class Chat {
     {
       systemMessage = "",
       max_tokens = 4096,
+      max_gen_tokens = 2048,
+      enable_max_gen_tokens = true,
       tokens_margin = 1024,
       apiEndPoint = "https://api.openai.com/v1/chat/completions",
       model = "gpt-3.5-turbo",
@@ -96,6 +100,8 @@ class Chat {
     this.messages = [];
     this.total_tokens = calculate_token_length(systemMessage);
     this.max_tokens = max_tokens;
+    this.max_gen_tokens = max_gen_tokens;
+    this.enable_max_gen_tokens = enable_max_gen_tokens;
     this.tokens_margin = tokens_margin;
     this.sysMessageContent = systemMessage;
     this.apiEndpoint = apiEndPoint;
@@ -151,6 +157,9 @@ class Chat {
     if (this.enable_top_p) {
       body["top_p"] = this.top_p;
     }
+    if (this.enable_max_gen_tokens) {
+      body["max_tokens"] = this.max_gen_tokens;
+    }
 
     return fetch(this.apiEndpoint, {
       method: "POST",

src/settings.tsx

@@ -188,6 +188,7 @@ const Number = (props: {
   field:
     | "totalTokens"
     | "maxTokens"
+    | "maxGenTokens"
     | "tokenMargin"
     | "postBeginIndex"
     | "presence_penalty"
@@ -197,9 +198,27 @@ const Number = (props: {
 }) => {
   return (
     <Help help={props.help}>
-      <label className="m-2 p-2">{props.field}</label>
+      <span>
+        <label className="m-2 p-2">{props.field}</label>
+        {props.field === "maxGenTokens" && (
+          <input
+            type="checkbox"
+            checked={props.chatStore.maxGenTokens_enabled}
+            onChange={() => {
+              const newChatStore = { ...props.chatStore };
+              newChatStore.maxGenTokens_enabled =
+                !newChatStore.maxGenTokens_enabled;
+              props.setChatStore({ ...newChatStore });
+            }}
+          />
+        )}
+      </span>
       <input
         readOnly={props.readOnly}
+        disabled={
+          props.field === "maxGenTokens" &&
+          !props.chatStore.maxGenTokens_enabled
+        }
         type="number"
         className="m-2 p-2 border rounded focus w-28"
         value={props.chatStore[props.field]}
@@ -384,6 +403,12 @@ export default (props: {
             readOnly={false}
             {...props}
           />
+          <Number
+            field="maxGenTokens"
+            help="最大生成 Tokens 数量"
+            readOnly={false}
+            {...props}
+          />
           <Number
             field="tokenMargin"
             help="当 totalTokens > maxTokens - tokenMargin 时会触发历史消息裁切，chatgpt会“忘记”一部分对话中的消息（但所有历史消息仍然保存在本地）"