使用Retrofit调用OpenAI API无法获取REST结果求助

我搜索许久未找到有效解决方案，调用OpenAI API时持续报错，报错提示为Invalid model: text-davinci-003。相关代码如下：

RetrofitInstance #

class RetrofitInstance {

    companion object{
        private lateinit var RETROFIT: Retrofit
        private var client = OkHttpClient.Builder().apply {
            addInterceptor(ChatGPTInterceptor())
        }.build()

        fun getRetrofitInstance(pathUrl: String): Retrofit{
            if(!::RETROFIT.isInitialized){
                RETROFIT = Retrofit.Builder()
                    .baseUrl(pathUrl)
                    .client(client)
                    .addConverterFactory(GsonConverterFactory.create())
                    .build()
            }
            return RETROFIT
        }
    }

}

ChatGPTInterceptor #

class ChatGPTInterceptor: Interceptor {
    override fun intercept(chain: Interceptor.Chain): Response {
        val proceed = chain.request()
            .newBuilder()
            .addHeader("content-type", "application/json")
            .addHeader("X-RapidAPI-Key", "Key-Censored")
            .addHeader("X-RapidAPI-Host", "openai80.p.rapidapi.com")
            .build()
        return chain.proceed(proceed)
    }
}

MainActivity #

val sendMessageModel = SendMessageModel("user", binding.inputText.text.toString())

val realSendMessageModel = RealSendMessageModel("text-davinci-003", arrayListOf(sendMessageModel))

val retrofitInstance = RetrofitInstance.getRetrofitInstance("https://openai80.p.rapidapi.com/")
val endpoint = retrofitInstance.create(ChatGPTEndpoint::class.java)
val callback = endpoint.getRetrofitAnswer(realSendMessageModel)
callback.enqueue(object: Callback<AnswerModel>{
    override fun onResponse(call: Call<AnswerModel>, response: Response<AnswerModel>) {
        val s = response
    }

    override fun onFailure(call: Call<AnswerModel>, t: Throwable) {
        val s = t
    }
})

ChatGPTEndpoint #

interface ChatGPTEndpoint {

    @POST("chat/completions")
    fun getRetrofitAnswer(@Body message: RealSendMessageModel): Call<AnswerModel>

}

排查及修复方案 #

• 模型与接口不匹配：chat/completions接口仅支持GPT-3.5/4系列聊天模型（如gpt-3.5-turbo），text-davinci-003是文本补全模型，对应completions接口。要么将接口路径改为completions，要么把模型参数换成gpt-3.5-turbo。
• Retrofit单例逻辑错误：当前单例实现会导致第一次初始化后，后续传入不同baseUrl也会复用旧实例，可能引发地址错误。建议去掉初始化判断，每次调用创建新实例，或者按baseUrl缓存不同实例。
• 错误信息捕获不全：当前onResponse未处理非成功响应，建议添加错误体打印，获取详细报错信息：

  override fun onResponse(call: Call<AnswerModel>, response: Response<AnswerModel>) {
      if (response.isSuccessful) {
          // 处理成功结果
      } else {
          val errorMsg = response.errorBody()?.string() ?: "未知错误"
          Log.e("API_ERROR", errorMsg)
      }
  }
  

• 验证RapidAPI配置：确认X-RapidAPI-Key有效，且订阅权限覆盖目标接口，同时检查后台是否有请求超限等限制。

内容的提问来源于stack exchange，提问作者Antony Lajes