support ChatGLM_6b chat stream

Author: little51

ChatGLM_6b.py

@@ -1,8 +1,12 @@
 import requests
 import json
+import os
 
 def getAnswerFromChatGLM6b(context):
-    url = 'http://172.16.62.136:8000/'
+    if os.name == 'nt':
+        url = 'http://172.16.62.66:8000/stream'
+    else:
+        url = 'http://172.16.62.136:8000/stream'
     data = '{"prompt": "'  + context +  '", "history": []}'
     headers = {'content-type': 'application/json;charset=utf-8'}
     r = requests.post(url, data=data.encode(), headers=headers)

chat/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chat",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "private": true,
   "dependencies": {
     "@chatui/core": "^2.4.2",

chat/src/App.js

@@ -30,7 +30,7 @@ const defaultQuickReplies = [
   {
     icon: 'message',
     name: 'golang',
-  },  
+  },
 ];
 
 
@@ -52,10 +52,10 @@ function App() {
   const { messages, appendMsg, setTyping } = useMessages(initialMessages);
   const [percentage, setPercentage] = useState(0);
 
-  function handleSend(type, val) {
-    if (percentage>0) {
-      alert("正在生成，请稍候！") ;
-      return ;
+  function handleSend(type, val, item_name) {
+    if (percentage > 0) {
+      alert("正在生成，请稍候！");
+      return;
     }
     if (type === 'text' && val.trim()) {
       appendMsg({
@@ -64,10 +64,14 @@ function App() {
         position: 'left',
         user: { avatar: '//gitclone.com/download1/user.png' },
       });
-
       setTyping(true);
       setPercentage(10);
-      onGenCode(val, val, 0);
+      if (item_name === undefined) {
+        if (isChinese(val)) {
+          item_name = "GPT";
+        }
+      }
+      onGenCode(val, val, 0, item_name);
     }
   }
 
@@ -103,23 +107,38 @@ function App() {
     } else {
       content = "写一个python版的数组排序";
     }
-    handleSend('text', content);
+    handleSend('text', content, item.name);
+  }
+
+  function Sleep(ms) {
+    return new Promise(resolve => setTimeout(resolve, ms))
   }
 
-  function onGenCode(context_en, context_ch, count) {
-    if (count >= 5) {
+  async function onGenCode(context_en, context_ch, count, item_name) {
+    var context_gpt = context_en;
+    var stop = false;
+    var x = 5;
+    if (item_name === "GPT") {
+      x = 1000;
+      await Sleep(500);
+    }
+    if (count >= x) {
       setPercentage(0);
       return;
     }
     let xhr = new XMLHttpRequest();
     xhr.open('post', 'https://gitclone.com/aiit/codegen_stream');
+    //xhr.open('post', 'http://localhost:5000/codegen_stream');
     xhr.setRequestHeader('Content-Type', 'application/json');
     xhr.onload = function () {
       var json = JSON.parse(xhr.response);
       if (count === 0) {
         context_en = context_en + "\n" + json.result_en;
         context_ch = context_ch + "\n" + json.result_ch;
-        var stop = json.stop ;
+        stop = json.stop;
+        if (item_name === "GPT") {
+          context_ch = json.result_ch;
+        }
         appendMsg({
           type: 'text',
           content: { text: context_ch },
@@ -132,20 +151,34 @@ function App() {
         }
         context_en = context_en + json.result_en;
         context_ch = context_ch + json.result_ch;
+        stop = json.stop;
         if (context_ch === context_en) {
-          updateMsg(context_en);
+          if (item_name === "GPT") {
+            updateMsg(json.result_en);
+          }
+          else {
+            updateMsg(context_en);
+          }
         } else {
-          updateMsg(context_ch + "\n" + context_en);
+          if (item_name === "GPT") {
+            updateMsg(json.result_en);
+          } else {
+            updateMsg(context_ch + "\n" + context_en);
+          }
         }
 
       }
       count++;
       setPercentage(count * 20);
-      if(stop){
+      if (stop) {
         setPercentage(0);
         return;
-      } 
-      onGenCode(context_en, context_ch, count);
+      }
+      if (item_name === "GPT") {
+        onGenCode(context_gpt, context_gpt, count, item_name);
+      } else {
+        onGenCode(context_en, context_ch, count, item_name);
+      }
     }
     xhr.send(JSON.stringify({
       "context": context_en,
@@ -187,6 +220,15 @@ function App() {
     }
   }
 
+  function isChinese(s) {
+    let reg = new RegExp("[\\u4E00-\\u9FFF]+", "g")
+    if (reg.test(s)) {
+      return true;
+    } else {
+      return false;
+    }
+  }
+
   useEffect(() => {
     var oUl = document.getElementById('root');
     var aBox = getByClass(oUl, 'Input Input--outline Composer-input');

codegen.py

@@ -43,9 +43,10 @@ async def codegen(request):
     print(time.strftime("%Y-%m-%d %H:%M:%S",time.localtime()),"context : " + context)
     context = context.replace("//","").replace("#","").strip()
     stop = False
-    if flag_chs :#or content.startwith('gpt-j') :
-        # result = getAnswerFromChatGPTJ(context,maxlength).replace(context,"")
+    if flag_chs :
         result = getAnswerFromChatGLM6b(context)
+        stop = result.endswith("[stop]")
+        result = result.replace("[stop]", "")
     else:
         result,stop = sampling(context,maxlength)
     end = time.perf_counter()

codegen_stream.py

@@ -26,13 +26,12 @@ async def codegen_stream(request):
     flag_chs = f(context)
     stop = False
     if flag_chs:
-        # results = sampling_gptj(context, maxlength)
-        # results = json.loads(results)
-        # result_en = results["result_en"]
-        # result_ch = results["result_ch"]
         result_en = getAnswerFromChatGLM6b(context)
-        result_ch = result_en
-        stop = True
+        stop = result_en.endswith("[stop]")
+        result_ch = result_en.replace("[stop]", "")
+        if result_ch == "" :
+            result_ch = "思考中"
+        result_en = result_ch
     else:
         result_en,stop = sampling(context, maxlength)
         result_ch = result_en