feat: 支持非流式传输

2025-07-09 21:52:00 +08:00
parent 716c344780
commit c5a9b9a15f
9 changed files with 365 additions and 48 deletions
--- a/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application.Contracts/Dtos/OpenAi/ChatCompletionsOutput.cs
+++ b/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application.Contracts/Dtos/OpenAi/ChatCompletionsOutput.cs
@@ -0,0 +1,115 @@
+using Newtonsoft.Json;
+
+namespace Yi.Framework.AiHub.Application.Contracts.Dtos.OpenAi;
+
+public class ChatCompletionsOutput
+{
+    [JsonProperty("id")]
+    public string Id { get; set; }
+    [JsonProperty("object")]
+    public string Object { get; set; }
+    [JsonProperty("created_at")]
+    public long CreatedAt { get; set; }
+    [JsonProperty("status")]
+    public string Status { get; set; }
+    [JsonProperty("error")]
+    public object Error { get; set; }
+    [JsonProperty("incomplete_details")]
+    public object IncompleteDetails { get; set; }
+    [JsonProperty("instructions")]
+    public object Instructions { get; set; }
+    [JsonProperty("max_output_tokens")]
+    public object MaxOutputTokens { get; set; }
+    [JsonProperty("model")]
+    public string Model { get; set; }
+    [JsonProperty("output")]
+    public List<Output> Output { get; set; }
+    [JsonProperty("parallel_tool_calls")]
+    public bool ParallelToolCalls { get; set; }
+    [JsonProperty("previous_response_id")]
+    public object PreviousResponseId { get; set; }
+    [JsonProperty("reasoning")]
+    public Reasoning Reasoning { get; set; }
+    [JsonProperty("store")]
+    public bool Store { get; set; }
+    [JsonProperty("temperature")]
+    public double Temperature { get; set; }
+    [JsonProperty("text")]
+    public Text Text { get; set; }
+    [JsonProperty("tool_choice")]
+    public string ToolChoice { get; set; }
+    [JsonProperty("tools")]
+    public List<object> Tools { get; set; }
+    [JsonProperty("top_p")]
+    public double TopP { get; set; }
+    [JsonProperty("truncation")]
+    public string Truncation { get; set; }
+    [JsonProperty("usage")]
+    public Usage Usage { get; set; }
+    [JsonProperty("user")]
+    public object User { get; set; }
+    [JsonProperty("metadata")]
+    public Dictionary<string, object> Metadata { get; set; }
+}
+public class Output
+{
+    [JsonProperty("type")]
+    public string Type { get; set; }
+    [JsonProperty("id")]
+    public string Id { get; set; }
+    [JsonProperty("status")]
+    public string Status { get; set; }
+    [JsonProperty("role")]
+    public string Role { get; set; }
+    [JsonProperty("content")]
+    public List<Content> Content { get; set; }
+}
+public class Content
+{
+    [JsonProperty("type")]
+    public string Type { get; set; }
+    [JsonProperty("text")]
+    public string Text { get; set; }
+    [JsonProperty("annotations")]
+    public List<object> Annotations { get; set; }
+}
+public class Reasoning
+{
+    [JsonProperty("effort")]
+    public object Effort { get; set; }
+    [JsonProperty("summary")]
+    public object Summary { get; set; }
+}
+public class Text
+{
+    [JsonProperty("format")]
+    public Format Format { get; set; }
+}
+public class Format
+{
+    [JsonProperty("type")]
+    public string Type { get; set; }
+}
+public class Usage
+{
+    [JsonProperty("input_tokens")]
+    public int InputTokens { get; set; }
+    [JsonProperty("input_tokens_details")]
+    public InputTokensDetails InputTokensDetails { get; set; }
+    [JsonProperty("output_tokens")]
+    public int OutputTokens { get; set; }
+    [JsonProperty("output_tokens_details")]
+    public OutputTokensDetails OutputTokensDetails { get; set; }
+    [JsonProperty("total_tokens")]
+    public int TotalTokens { get; set; }
+}
+public class InputTokensDetails
+{
+    [JsonProperty("cached_tokens")]
+    public int CachedTokens { get; set; }
+}
+public class OutputTokensDetails
+{
+    [JsonProperty("reasoning_tokens")]
+    public int ReasoningTokens { get; set; }
+}
--- a/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application.Contracts/Dtos/SendMessageStreamOutputDto.cs
+++ b/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application.Contracts/Dtos/SendMessageStreamOutputDto.cs
@@ -1,6 +1,6 @@
 namespace Yi.Framework.AiHub.Application.Contracts.Dtos;

-public class SendMessageOutputDto
+public class SendMessageStreamOutputDto
 {
    /// <summary>
    /// 唯一标识符
--- a/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application/Services/OpenApiService.cs
+++ b/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Application/Services/OpenApiService.cs
@@ -55,6 +55,7 @@ public class OpenApiService : ApplicationService
            }
        }

+        //是否使用流式传输
        if (input.Stream)
        {
            //ai网关代理httpcontext
@@ -64,8 +65,9 @@ public class OpenApiService : ApplicationService
        }
        else
        {
-           await _aiGateWayManager.CompleteChatForStatisticsAsync(_httpContextAccessor.HttpContext,input.Model, history, userId, null,
-                    cancellationToken);
+            await _aiGateWayManager.CompleteChatForStatisticsAsync(_httpContextAccessor.HttpContext, input.Model,
+                history, userId, null,
+                cancellationToken);
        }
    }

--- a/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Domain/AiChat/Impl/AzureRestChatService.cs
+++ b/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Domain/AiChat/Impl/AzureRestChatService.cs
@@ -1,4 +1,5 @@
-using System.Runtime.CompilerServices;
+using System.Net.Http.Json;
+using System.Runtime.CompilerServices;
 using System.Text;
 using Newtonsoft.Json;
 using Newtonsoft.Json.Linq;
@@ -62,7 +63,8 @@ public class AzureRestChatService : IChatService

        if (!response.IsSuccessStatusCode)
        {
-            throw new UserFriendlyException($"当前模型不可用：{aiModelDescribe.ModelId}，状态码:{response.StatusCode},原因：{response.ReasonPhrase}");
+            throw new UserFriendlyException(
+                $"当前模型不可用：{aiModelDescribe.ModelId}，状态码:{response.StatusCode},原因：{response.ReasonPhrase}");
        }
        // 确认响应成功
        // response.EnsureSuccessStatusCode();
@@ -76,11 +78,11 @@ public class AzureRestChatService : IChatService
            var result = new CompleteChatResponse();
            try
            {
-                var jsonObj = MapToJObject(line);
+                var jsonObj = MapToStreamJObject(line);
                if (jsonObj is not null)
                {
-                    var content = GetContent(jsonObj);
-                    var tokenUsage = GetTokenUsage(jsonObj);
+                    var content = GetStreamContent(jsonObj);
+                    var tokenUsage = GetStreamTokenUsage(jsonObj);
                    result = new CompleteChatResponse
                    {
                        TokenUsage = tokenUsage,
@@ -98,28 +100,85 @@ public class AzureRestChatService : IChatService
        }
    }

-    public Task<CompleteChatResponse> CompleteChatAsync(AiModelDescribe aiModelDescribe, List<ChatMessage> messages, CancellationToken cancellationToken)
+    public async Task<CompleteChatResponse> CompleteChatAsync(AiModelDescribe aiModelDescribe,
+        List<ChatMessage> messages, CancellationToken cancellationToken)
    {
-        throw new NotImplementedException("暂未实现");
+        // 设置API URL
+        var apiUrl = $"{aiModelDescribe.Endpoint}";
+
+        // 准备请求内容
+        var requestBody = new
+        {
+            messages = messages.Select(x => new
+            {
+                role = x.GetRoleAsString(),
+                content = x.Content.FirstOrDefault()?.Text
+            }).ToList(),
+            stream = false,
+            // max_tokens = 2048,
+            // temperature = 0.8,
+            // top_p = 0.1,
+            // presence_penalty = 0,
+            // frequency_penalty = 0,
+            model = aiModelDescribe.ModelId
+        };
+
+        // 序列化请求内容为JSON
+        string jsonBody = JsonConvert.SerializeObject(requestBody);
+
+        using var httpClient = new HttpClient()
+        {
+            //10分钟超时
+            Timeout = TimeSpan.FromSeconds(600)
+        };
+        // 设置请求头
+        httpClient.DefaultRequestHeaders.Add("Authorization", $"Bearer {aiModelDescribe.ApiKey}");
+        // 其他头信息如Content-Type在StringContent中设置
+
+        // 构造 POST 请求
+        var request = new HttpRequestMessage(HttpMethod.Post, apiUrl);
+
+        // 设置请求内容（示例）
+        request.Content = new StringContent(jsonBody, Encoding.UTF8, "application/json");
+
+        // 发送POST请求
+        HttpResponseMessage response =
+            await httpClient.SendAsync(request, cancellationToken);
+
+        if (!response.IsSuccessStatusCode)
+        {
+            throw new UserFriendlyException(
+                $"当前模型不可用：{aiModelDescribe.ModelId}，状态码:{response.StatusCode},原因：{response.ReasonPhrase}");
+        }
+        // 确认响应成功
+        // response.EnsureSuccessStatusCode();
+
+        // 读取响应内容
+        var responseStr = await response.Content.ReadAsStringAsync(cancellationToken);
+        var jObject = MapToJObject(responseStr);
+
+        var content = GetContent(jObject);
+        var usage = GetTokenUsage(jObject);
+        var result = new CompleteChatResponse
+        {
+            TokenUsage = usage,
+            IsFinish = true,
+            Content = content
+        };
+
+        return result;
    }

-    private JObject? MapToJObject(string line)
+    private JObject? MapToJObject(string body)
    {
-        if (line == "data: [DONE]"||string.IsNullOrWhiteSpace(line) )
-        {
+        if (string.IsNullOrWhiteSpace(body))
            return null;
-        }
-
-        if (string.IsNullOrWhiteSpace(line))
-            return null;
-        string prefix = "data: ";
-        line = line.Substring(prefix.Length);
-        return JObject.Parse(line);
+        return JObject.Parse(body);
    }

    private string? GetContent(JObject? jsonObj)
    {
-        var contentToken = jsonObj.SelectToken("choices[0].delta.content");
+        var contentToken = jsonObj.SelectToken("choices[0].message.content");
        if (contentToken != null && contentToken.Type != JTokenType.Null)
        {
            return contentToken.ToString();
@@ -157,4 +216,60 @@ public class AzureRestChatService : IChatService

        return null;
    }
+
+
+    private JObject? MapToStreamJObject(string line)
+    {
+        if (line == "data: [DONE]" || string.IsNullOrWhiteSpace(line))
+        {
+            return null;
+        }
+
+        if (string.IsNullOrWhiteSpace(line))
+            return null;
+        string prefix = "data: ";
+        line = line.Substring(prefix.Length);
+        return JObject.Parse(line);
+    }
+
+    private string? GetStreamContent(JObject? jsonObj)
+    {
+        var contentToken = jsonObj.SelectToken("choices[0].delta.content");
+        if (contentToken != null && contentToken.Type != JTokenType.Null)
+        {
+            return contentToken.ToString();
+        }
+
+        return null;
+    }
+
+    private TokenUsage? GetStreamTokenUsage(JObject? jsonObj)
+    {
+        var usage = jsonObj.SelectToken("usage");
+        if (usage is not null && usage.Type != JTokenType.Null)
+        {
+            var result = new TokenUsage();
+            var completionTokens = usage["completion_tokens"];
+            if (completionTokens is not null && completionTokens.Type != JTokenType.Null)
+            {
+                result.OutputTokenCount = completionTokens.ToObject<int>();
+            }
+
+            var promptTokens = usage["prompt_tokens"];
+            if (promptTokens is not null && promptTokens.Type != JTokenType.Null)
+            {
+                result.InputTokenCount = promptTokens.ToObject<int>();
+            }
+
+            var totalTokens = usage["total_tokens"];
+            if (totalTokens is not null && totalTokens.Type != JTokenType.Null)
+            {
+                result.TotalTokenCount = totalTokens.ToObject<int>();
+            }
+
+            return result;
+        }
+
+        return null;
+    }
 }
--- a/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Domain/Managers/AiGateWayManager.cs
+++ b/Yi.Abp.Net8/module/ai-hub/Yi.Framework.AiHub.Domain/Managers/AiGateWayManager.cs
@@ -9,11 +9,13 @@ using Newtonsoft.Json.Serialization;
 using OpenAI.Chat;
 using Volo.Abp.Domain.Services;
 using Yi.Framework.AiHub.Application.Contracts.Dtos;
+using Yi.Framework.AiHub.Application.Contracts.Dtos.OpenAi;
 using Yi.Framework.AiHub.Domain.AiChat;
 using Yi.Framework.AiHub.Domain.Entities;
 using Yi.Framework.AiHub.Domain.Entities.Model;
 using Yi.Framework.AiHub.Domain.Shared.Dtos;
 using Yi.Framework.SqlSugarCore.Abstractions;
+using Usage = Yi.Framework.AiHub.Application.Contracts.Dtos.Usage;

 namespace Yi.Framework.AiHub.Domain.Managers;

@@ -107,7 +109,7 @@ public class AiGateWayManager : DomainService
        await using var writer = new StreamWriter(response.Body, Encoding.UTF8, leaveOpen: true);
        var modelDescribe = await GetModelAsync(modelId);
        var chatService = LazyServiceProvider.GetRequiredKeyedService<IChatService>(modelDescribe.HandlerName);
-        var output = await chatService.CompleteChatAsync(modelDescribe, messages, cancellationToken);
+        var data = await chatService.CompleteChatAsync(modelDescribe, messages, cancellationToken);
        if (userId is not null)
        {
            await _aiMessageManager.CreateUserMessageAsync(userId.Value, sessionId,
@@ -115,22 +117,23 @@ public class AiGateWayManager : DomainService
                {
                    Content = messages.LastOrDefault().Content.FirstOrDefault()?.Text ?? string.Empty,
                    ModelId = modelId,
-                    TokenUsage = output.TokenUsage,
+                    TokenUsage = data.TokenUsage,
                });

            await _aiMessageManager.CreateSystemMessageAsync(userId.Value, sessionId,
                new MessageInputDto
                {
-                    Content = output.Content,
+                    Content = data.Content,
                    ModelId = modelId,
-                    TokenUsage = output.TokenUsage
+                    TokenUsage = data.TokenUsage
                });

-            await _usageStatisticsManager.SetUsageAsync(userId.Value, modelId, output.TokenUsage.InputTokenCount,
-                output.TokenUsage.OutputTokenCount);
+            await _usageStatisticsManager.SetUsageAsync(userId.Value, modelId, data.TokenUsage.InputTokenCount,
+                data.TokenUsage.OutputTokenCount);
        }

-        var body = JsonConvert.SerializeObject(output, new JsonSerializerSettings
+        var result = MapToChatCompletions(modelId, data.Content);
+        var body = JsonConvert.SerializeObject(result, new JsonSerializerSettings
        {
            ContractResolver = new CamelCasePropertyNamesContractResolver()
        });
@@ -261,9 +264,9 @@ public class AiGateWayManager : DomainService
    }


-    private SendMessageOutputDto MapToMessage(string modelId, string content)
+    private SendMessageStreamOutputDto MapToMessage(string modelId, string content)
    {
-        var output = new SendMessageOutputDto
+        var output = new SendMessageStreamOutputDto
        {
            Id = "chatcmpl-BotYP3BlN5T4g9YPnW0fBSBvKzXdd",
            Object = "chat.completion.chunk",
@@ -338,4 +341,75 @@ public class AiGateWayManager : DomainService

        return output;
    }
+
+    private ChatCompletionsOutput MapToChatCompletions(string modelId, string content)
+    {
+        return new ChatCompletionsOutput
+        {
+            Id = "resp_67ccd2bed1ec8190b14f964abc0542670bb6a6b452d3795b",
+            Object = "response",
+            CreatedAt = 1741476542,
+            Status = "completed",
+            Error = null,
+            IncompleteDetails = null,
+            Instructions = null,
+            MaxOutputTokens = null,
+            Model = modelId,
+            Output = new List<Output>()
+            {
+                new Output
+                {
+                    Type = "message",
+                    Id = "msg_67ccd2bf17f0819081ff3bb2cf6508e60bb6a6b452d3795b",
+                    Status = "completed",
+                    Role = "assistant",
+                    Content = new List<Content>
+                    {
+                        new Content
+                        {
+                            Type = "output_text",
+                            Text = content,
+                            Annotations = new List<object>()
+                        }
+                    }
+                }
+            },
+            ParallelToolCalls = true,
+            PreviousResponseId = null,
+            Reasoning = new Reasoning
+            {
+                Effort = null,
+                Summary = null
+            },
+            Store = true,
+            Temperature = 0,
+            Text = new Text
+            {
+                Format = new Format
+                {
+                    Type = "text"
+                }
+            },
+            ToolChoice = "auto",
+            Tools = new List<object>(),
+            TopP = 1.0,
+            Truncation = "disabled",
+            Usage = new Application.Contracts.Dtos.OpenAi.Usage
+            {
+                InputTokens = 0,
+                InputTokensDetails = new InputTokensDetails
+                {
+                    CachedTokens = 0
+                },
+                OutputTokens = 0,
+                OutputTokensDetails = new OutputTokensDetails
+                {
+                    ReasoningTokens = 0
+                },
+                TotalTokens = 0
+            },
+            User = null,
+            Metadata = null
+        };
+    }
 }