java - OpenAI ChatGPT (GPT-3.5) API 错误 400 : "Unexpected response code 400 for https://api.openai.com/v1/completions" (migrating GPT-3 to GPT-3. 5 API)

Question

我有一个 Android 应用程序，目前正在使用 chat gpt 3.0 进行补全，并且工作正常。现在，在他们发布 chat gpt 3.5 Turbo 后，我根据他们的请求示例做了一些更改，但抛出了 400 错误，我感谢任何帮助，谢谢

• 我的 gpt 3.0 代码(工作正常)

  public static void getResponse(Context context, String URL, String Token, TextView mQuestionText, TextInputEditText queryEdt, TextView mResponseText, String query) throws JSONException {

        mQuestionText.setText(query);
        queryEdt.setText("");
        mResponseText.setText("Please wait..");

        RequestQueue requestQueue = Volley.newRequestQueue(context);

        JSONObject jsonObject = new JSONObject();
        jsonObject.put("model", "text-davinci-003");
        jsonObject.put("prompt", query);
        jsonObject.put("temperature", 0);
        jsonObject.put("max_tokens", 100);
        jsonObject.put("top_p", 1);
        jsonObject.put("frequency_penalty", 0.0);
        jsonObject.put("presence_penalty", 0.0);


        JsonObjectRequest jsonObjectRequest =  new JsonObjectRequest(Request.Method.POST, URL ,jsonObject, response -> {
            try {
                String responseMsg = response.getJSONArray("choices").getJSONObject(0).getString("text");
                mResponseText.setText(responseMsg);

                // SHUT DOWN TEXT TO SPEECH IN CASE OF QUERY CHANGE
                if(textToSpeech != null){
                    textToSpeech.stop();
                    textToSpeech.shutdown();
                    textToSpeech  = null;
                }

                // SPEAK THE RESPONSE FETCHED FROM SERVER
                textToSpeech(context,responseMsg);
            }catch (Exception e){
                // error
                Log.d("TAG","Error is   " + e.getMessage());
            }
        }, error -> {
            Log.d("TAG","Error " + error.getMessage());
        }){
            @Override
            public Map<String, String> getHeaders() {
                Map<String,String> params = new HashMap<>();
                params.put("Content-Type","application/json");
                params.put("Authorization","Bearer " + Token);
                return params;
            }
        };

        requestQueue.add(jsonObjectRequest);
    }

• 现在切换到 3.5 Turbo，我使用 gpt-3.5-turbo 作为模型

 public static void getResponse(Context context, String URL, String Token, TextView mQuestionText, TextInputEditText queryEdt, TextView mResponseText, String query) throws JSONException {

        mQuestionText.setText(query);
        queryEdt.setText("");
        mResponseText.setText("Please wait..");

        RequestQueue requestQueue = Volley.newRequestQueue(context);

        ArrayList<ChatModel> arrayList = new ArrayList<>();
        arrayList.add(new ChatModel("user",query));

        JSONObject jsonObject = new JSONObject();
        jsonObject.put("model", "gpt-3.5-turbo");
        jsonObject.put("messages",arrayList);
        

        JsonObjectRequest jsonObjectRequest =  new JsonObjectRequest(Request.Method.POST, URL ,jsonObject, response -> {
            try {
                String responseMsg = response.getJSONArray("choices").getJSONObject(0).getString("text");
                mResponseText.setText(responseMsg);

                // SHUT DOWN TEXT TO SPEECH IN CASE OF QUERY CHANGE
                if(textToSpeech != null){
                    textToSpeech.stop();
                    textToSpeech.shutdown();
                    textToSpeech  = null;
                }

                // SPEAK THE RESPONSE FETCHED FROM SERVER
                textToSpeech(context,responseMsg);
            }catch (Exception e){
                // error
                Log.d("TAG","Error is   " + e.getMessage());
            }
        }, error -> {
            Log.d("TAG","Error " + error.getMessage());
        }){
            @Override
            public Map<String, String> getHeaders() {
                Map<String,String> params = new HashMap<>();
                params.put("Content-Type","application/json");
                params.put("Authorization","Bearer " + Token);
                return params;
            }
        };
        requestQueue.add(jsonObjectRequest);
    }

• 使用 chat gpt 3.5 Turbo 模型时出错(当我使用 chat gpt 3.0 时它可以工作)

  E/Volley: [1922] NetworkUtility.shouldRetryException: Unexpected response code 400 for 
  https://api.openai.com/v1/completions

• 基于他们的文档

Answer 1

您使用的是 gpt-3.5-turbo 型号。

Chat Completions API 之间存在三个主要区别(即 GPT-3.5 API)和 Completions API (即 GPT-3 API):

1. API端点
   • Completions API:https://api.openai.com/v1/completions
   • 聊天完成 API:https://api.openai.com/v1/chat/completions
2. prompt 参数(Completions API)已替换为 messages 参数(Chat Completions API)
3. 响应访问
   • Completions API:
     response.getJSONArray("choices").getJSONObject(0).getString("text")
   • 聊天完成 API:response.getJSONArray("choices").getJSONObject(0).getString("message")

问题 1:您使用了错误的 API 端点

改变这个...

https://api.openai.com/v1/completions

...对此。

https://api.openai.com/v1/chat/completions

问题 2:确保 messages 参数的 JSON 有效

像这样:

JSONObject jsonObject = new JSONObject();
JSONArray messages = new JSONArray();
JSONObject message = new JSONObject();

message.put("role", "user");
message.put("content", "What are your functionalities?");

messages.put(message);

jsonObject.put("messages", messages);

此代码将创建以下 JSON 对象:

{
  "messages": [
    {
      "role": "user",
      "content": "What are your functionalities?"
    }
  ]
}

问题 3:您访问响应的方式不正确

改变这个...

response.getJSONArray("choices").getJSONObject(0).getString("text")

...对此。

response.getJSONArray("choices").getJSONObject(0).getString("message")

注意:如果您仍然收到 API 错误 400，请确保您已将 Content-Type header 设置为 application/json。请注意，application/json、UTF-8 不起作用。