2. 技术交流
  3. 在处理gpt返回数据时需要 处理的流式问题疑问？

在处理gpt返回数据时需要 处理的流式问题疑问？

aardio 2023-12-13 1057

现在 gpt 返回的数据都是流的形式，怎么样才能把它转换为类似正常的一串字符串呢？请教，下面内容 ，实际我要取的是 response 对应 的内容

Code HTML/XTMLLine:24复制
	
  • 1.
  • 2.
  • 3.
  • 4.
  • 5.
  • 6.
  • 7.
  • 8.
  • 9.
  • 10.
  • 11.
  • 12.
  • 13.
  • 14.
  • 15.
  • 16.
  • 17.
  • 18.
  • 19.
  • 20.
  • 21.
  • 22.
  • 23.
  • 24.
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.330511Z","response":"\n","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.398591Z","response":"G","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.475653Z","response":"PT","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.566181Z","response":":","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.630396Z","response":" Hello","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.70039Z","response":"!","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.773188Z","response":" It","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.841229Z","response":"'","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.904017Z","response":"s","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:44.971747Z","response":" nice","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.035208Z","response":" to","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.099293Z","response":" meet","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.165876Z","response":" you","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.22995Z","response":" too","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.291708Z","response":".","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.35596Z","response":" What","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.418584Z","response":" can","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.479481Z","response":" I","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.546164Z","response":" help","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.614087Z","response":" you","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.668333Z","response":" with","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.729018Z","response":" today","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.798808Z","response":"?","done":false}
    • {"model":"vicuna","created_at":"2023-12-13T09:44:45.856087Z","response":"","done":true,"context":[319,13563,1546,263,12758,1404,322,385,23116,21082,20255,29889,450,20255,4076,8444,29892,13173,29892,322,1248,568,6089,304,278,1404,29915,29879,5155,29889,13,13,11889,29901,20103,304,5870,366,29973,13,22933,9047,13566,29901,13,13,29954,7982,29901,15043,29991,739,29915,29879,7575,304,5870,366,2086,29889,1724,508,306,1371,366,411,9826,29973],"total_duration":3704897916,"prompt_eval_count":46,"prompt_eval_duration":2217278000,"eval_count":23,"eval_duration":1462403000}


    最新回复 (7)
    • aardio 2023-12-13
      0 2楼
      生成 这样一句" Hello! It's nice to meet you too. What can I help you with today?
    • aardio 2023-12-13
      0 3楼

      尝试了一下，大概的方法是先将内容按换行拆，再提取内容组合起来，利用正则提取关键的 2 组数据

      Code AardioLine:2复制
      	
    • 1.
    • 2.
      • preg('"created_at":"(.*?)","response":"(.*?)"');
      • regex.gmatch( test  )

      但是这种方法并没有对时间进行排序。。。

    • 小光芒 2023-12-13
      0 4楼

      这个玩意不是json吗，转table直接用呗

    • aardio 2023-12-14
      0 5楼
      嘿嘿嘿。。。合起来不是 Json
    • 瞌睡蟲子 2023-12-15
      0 6楼

      如果是http流式请求，可以在onReceive里面处理，每一条是一个json。

      Code AardioLine:24复制
      	
    • 1.
    • 2.
    • 3.
    • 4.
    • 5.
    • 6.
    • 7.
    • 8.
    • 9.
    • 10.
    • 11.
    • 12.
    • 13.
    • 14.
    • 15.
    • 16.
    • 17.
    • 18.
    • 19.
    • 20.
    • 21.
    • 22.
    • 23.
    • 24.
      • // http rest客户端
      • var http = web.rest.jsonClient(); 
      • var api = http.api("https://xxx.xxx.com/"); 
      • http.beforeRequestHeaders = function(params){    
      •     return {"Authorization":"Bearer xxx"}; 
      • }
      • http._http.onReceive = function(str,size,contentLength){
      •     if(str["data"]["choices"]){
      •         for(k,v in str["data"]["choices"]){
      •             lastMsg += v["delta"]["content"];
      •             mainForm.edit_log.appendText(v["delta"]["content"]);        
      •             //mainForm.edit_log.print(v["delta"]["content"]);
      •         }
      •     }else {        
      •         table.insert(history,{ "role": "assistant", "content": lastMsg });
      •         if(#history > msgSize){
      •             table.pop(history);
      •         }
      •         lastMsg = "";
      •         mainForm.edit_log.print("");
      •     }    
      • }


    • aardio 2023-12-15
      0 7楼
      瞌睡蟲子 如果是http流式请求，可以在onReceive里面处理，每一条是一个json。// http rest客户端 var http&nbs ...
      十分感谢。。。。我还在想如果返回的数据量大，后续的处理就成问题了 代码 让我想起了以前aardio 作者写的机器人。。。
    • aardio 2023-12-16
      0 8楼
      "stream": false 可以避免返回数据的问题
    返回
    发新帖
    作者最近主题：