模拟 OpenAI 聊天客户端

Question

如何使用 MOQ 来模拟 OpenAIClient 以返回完成响应？我有一个 json 响应，想将其反序列化为

Response<Completions>

。

      private OpenAIClient client = // initialize;
    
      private async Task<Response<Completions>> Complete(string model, string prompt) =>
          await client.GetCompletionsAsync(model, prompt);

以下返回响应的尝试失败，因为 Completions 类不包含可用于反序列化 json 响应的构造函数 -

private readonly Mock<OpenAIClient> client= new();
          client
              .Setup(x => x.GetCompletionsAsync(It.IsAny<string>(), It.IsAny<string>(), It.IsAny<CancellationToken>()))
              .ReturnsAsync(Response.FromValue(JsonConvert.DeserializeObject<Completions>(completionsJson), Mock.Of<Response>()));

Completions 类构造函数标记为

internal

，Completions 类中的所有对象也是如此。

Answer 1

我知道这是一个老问题，但我遇到了同样的问题，并通过创建一个由替代品/模拟（我正在使用

NSubstitute

）与真实对象混合组成的弗兰肯斯坦对象来解决它。

const string userPrompt = "test";
const string responsePrompt = "response";

// real completion
var completion = OpenAIChatModelFactory.ChatCompletion(
    role: ChatMessageRole.User,
    content: new ChatMessageContent(ChatMessageContentPart.CreateTextPart(responsePrompt)));

// mocked PipelineResponse
var pipelineResponse = Substitute.For<PipelineResponse>();
// real ClientResult
var completionResponse = ClientResult<ChatCompletion>.FromValue(completion, pipelineResponse);

// mocking my final result
_chatClientMock.CompleteChatAsync(Arg.Any<List<ChatMessage>>(), Arg.Any<ChatCompletionOptions>(), Arg.Any<CancellationToken>())
    .Returns(completionResponse);
// call your test service here that invokes _chatClientMock.CompleteChatAsync 
// assertions...