feat: 支持Claude模型API类型及尊享包校验与扣减逻辑
This commit is contained in:
@@ -16,6 +16,7 @@ using Yi.Framework.AiHub.Domain.Entities;
|
||||
using Yi.Framework.AiHub.Domain.Entities.Model;
|
||||
using Yi.Framework.AiHub.Domain.Extensions;
|
||||
using Yi.Framework.AiHub.Domain.Managers;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Consts;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.OpenAi;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Enums;
|
||||
@@ -35,17 +36,19 @@ public class AiChatService : ApplicationService
|
||||
private readonly AiBlacklistManager _aiBlacklistManager;
|
||||
private readonly ILogger<AiChatService> _logger;
|
||||
private readonly AiGateWayManager _aiGateWayManager;
|
||||
private readonly PremiumPackageManager _premiumPackageManager;
|
||||
|
||||
public AiChatService(IHttpContextAccessor httpContextAccessor,
|
||||
AiBlacklistManager aiBlacklistManager,
|
||||
ISqlSugarRepository<AiModelEntity> aiModelRepository,
|
||||
ILogger<AiChatService> logger, AiGateWayManager aiGateWayManager)
|
||||
ILogger<AiChatService> logger, AiGateWayManager aiGateWayManager, PremiumPackageManager premiumPackageManager)
|
||||
{
|
||||
_httpContextAccessor = httpContextAccessor;
|
||||
_aiBlacklistManager = aiBlacklistManager;
|
||||
_aiModelRepository = aiModelRepository;
|
||||
_logger = logger;
|
||||
_aiGateWayManager = aiGateWayManager;
|
||||
_premiumPackageManager = premiumPackageManager;
|
||||
}
|
||||
|
||||
|
||||
@@ -118,6 +121,17 @@ public class AiChatService : ApplicationService
|
||||
}
|
||||
}
|
||||
|
||||
//如果是尊享包服务,需要校验是是否尊享包足够
|
||||
if (CurrentUser.IsAuthenticated && PremiumPackageConst.ModeIds.Contains(input.Model))
|
||||
{
|
||||
// 检查尊享token包用量
|
||||
var availableTokens = await _premiumPackageManager.GetAvailableTokensAsync(CurrentUser.GetId());
|
||||
if (availableTokens <= 0)
|
||||
{
|
||||
throw new UserFriendlyException("尊享token包用量不足,请先购买尊享token包");
|
||||
}
|
||||
}
|
||||
|
||||
//ai网关代理httpcontext
|
||||
await _aiGateWayManager.CompleteChatStreamForStatisticsAsync(_httpContextAccessor.HttpContext, input,
|
||||
CurrentUser.Id, sessionId, cancellationToken);
|
||||
@@ -154,7 +168,7 @@ public class AiChatService : ApplicationService
|
||||
{
|
||||
input.Model = "DeepSeek-R1-0528";
|
||||
}
|
||||
|
||||
|
||||
//ai网关代理httpcontext
|
||||
await _aiGateWayManager.CompleteChatStreamForStatisticsAsync(_httpContextAccessor.HttpContext, input,
|
||||
CurrentUser.Id, null, cancellationToken);
|
||||
|
||||
@@ -57,6 +57,18 @@ public class OpenApiService : ApplicationService
|
||||
var httpContext = this._httpContextAccessor.HttpContext;
|
||||
var userId = await _tokenManager.GetUserIdAsync(GetTokenByHttpContext(httpContext));
|
||||
await _aiBlacklistManager.VerifiyAiBlacklist(userId);
|
||||
|
||||
//如果是尊享包服务,需要校验是是否尊享包足够
|
||||
if (PremiumPackageConst.ModeIds.Contains(input.Model))
|
||||
{
|
||||
// 检查尊享token包用量
|
||||
var availableTokens = await _premiumPackageManager.GetAvailableTokensAsync(userId);
|
||||
if (availableTokens <= 0)
|
||||
{
|
||||
throw new UserFriendlyException("尊享token包用量不足,请先购买尊享token包");
|
||||
}
|
||||
}
|
||||
|
||||
//ai网关代理httpcontext
|
||||
if (input.Stream == true)
|
||||
{
|
||||
|
||||
@@ -0,0 +1,8 @@
|
||||
using Microsoft.AspNetCore.Mvc.ModelBinding.Metadata;
|
||||
|
||||
namespace Yi.Framework.AiHub.Domain.Shared.Consts;
|
||||
|
||||
public class PremiumPackageConst
|
||||
{
|
||||
public static List<string> ModeIds = ["claude-sonnet-4-5-20250929"];
|
||||
}
|
||||
@@ -0,0 +1,7 @@
|
||||
namespace Yi.Framework.AiHub.Domain.Shared.Enums;
|
||||
|
||||
public enum ModelApiTypeEnum
|
||||
{
|
||||
OpenAi,
|
||||
Claude
|
||||
}
|
||||
@@ -0,0 +1,861 @@
|
||||
using System.Diagnostics;
|
||||
using System.Net;
|
||||
using System.Net.Http.Json;
|
||||
using System.Text.Json;
|
||||
using Microsoft.Extensions.Logging;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.Anthropic;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.OpenAi;
|
||||
|
||||
namespace Yi.Framework.AiHub.Domain.AiGateWay.Impl.ThorClaude.Chats;
|
||||
|
||||
public sealed class ClaudiaChatCompletionsService(
|
||||
IHttpClientFactory httpClientFactory,
|
||||
ILogger<ClaudiaChatCompletionsService> logger)
|
||||
: IChatCompletionService
|
||||
{
|
||||
public List<ThorChatChoiceResponse> CreateResponse(AnthropicChatCompletionDto completionDto)
|
||||
{
|
||||
var response = new ThorChatChoiceResponse();
|
||||
var chatMessage = new ThorChatMessage();
|
||||
if (completionDto == null)
|
||||
{
|
||||
return new List<ThorChatChoiceResponse>();
|
||||
}
|
||||
|
||||
if (completionDto.content.Any(x => x.type.Equals("thinking", StringComparison.OrdinalIgnoreCase)))
|
||||
{
|
||||
// 将推理字段合并到返回对象去
|
||||
chatMessage.ReasoningContent = completionDto.content
|
||||
.First(x => x.type.Equals("thinking", StringComparison.OrdinalIgnoreCase)).Thinking;
|
||||
|
||||
chatMessage.Role = completionDto.role;
|
||||
chatMessage.Content = completionDto.content
|
||||
.First(x => x.type.Equals("text", StringComparison.OrdinalIgnoreCase)).text;
|
||||
}
|
||||
else
|
||||
{
|
||||
chatMessage.Role = completionDto.role;
|
||||
chatMessage.Content = completionDto.content
|
||||
.FirstOrDefault()?.text;
|
||||
}
|
||||
|
||||
response.Delta = chatMessage;
|
||||
response.Message = chatMessage;
|
||||
|
||||
if (completionDto.content.Any(x => x.type.Equals("tool_use", StringComparison.OrdinalIgnoreCase)))
|
||||
{
|
||||
var toolUse = completionDto.content
|
||||
.First(x => x.type.Equals("tool_use", StringComparison.OrdinalIgnoreCase));
|
||||
|
||||
chatMessage.ToolCalls =
|
||||
[
|
||||
new()
|
||||
{
|
||||
Id = toolUse.id,
|
||||
Function = new ThorChatMessageFunction()
|
||||
{
|
||||
Name = toolUse.name,
|
||||
Arguments = JsonSerializer.Serialize(toolUse.input,
|
||||
ThorJsonSerializer.DefaultOptions),
|
||||
},
|
||||
Index = 0,
|
||||
}
|
||||
];
|
||||
|
||||
return
|
||||
[
|
||||
response
|
||||
];
|
||||
}
|
||||
|
||||
return new List<ThorChatChoiceResponse> { response };
|
||||
}
|
||||
|
||||
private object CreateMessage(List<ThorChatMessage> messages, AiModelDescribe options)
|
||||
{
|
||||
var list = new List<object>();
|
||||
|
||||
foreach (var message in messages)
|
||||
{
|
||||
// 如果是图片
|
||||
if (message.ContentCalculated is IList<ThorChatMessageContent> contentCalculated)
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
role = message.Role,
|
||||
content = (List<object>)contentCalculated.Select<ThorChatMessageContent, object>(x =>
|
||||
{
|
||||
if (x.Type == "text")
|
||||
{
|
||||
if ("true".Equals(options.ModelExtraInfo, StringComparison.OrdinalIgnoreCase))
|
||||
{
|
||||
return new
|
||||
{
|
||||
type = "text",
|
||||
text = x.Text,
|
||||
cache_control = new
|
||||
{
|
||||
type = "ephemeral"
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
return new
|
||||
{
|
||||
type = "text",
|
||||
text = x.Text
|
||||
};
|
||||
}
|
||||
|
||||
var isBase64 = x.ImageUrl?.Url.StartsWith("http") == true;
|
||||
|
||||
if ("true".Equals(options.ModelExtraInfo, StringComparison.OrdinalIgnoreCase))
|
||||
{
|
||||
return new
|
||||
{
|
||||
type = "image",
|
||||
source = new
|
||||
{
|
||||
type = isBase64 ? "base64" : "url",
|
||||
media_type = "image/png",
|
||||
data = x.ImageUrl?.Url,
|
||||
},
|
||||
cache_control = new
|
||||
{
|
||||
type = "ephemeral"
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
return new
|
||||
{
|
||||
type = "image",
|
||||
source = new
|
||||
{
|
||||
type = isBase64 ? "base64" : "url",
|
||||
media_type = "image/png",
|
||||
data = x.ImageUrl?.Url,
|
||||
}
|
||||
};
|
||||
})
|
||||
});
|
||||
}
|
||||
else
|
||||
{
|
||||
if ("true".Equals(options.ModelExtraInfo, StringComparison.OrdinalIgnoreCase))
|
||||
{
|
||||
if (message.Role == "system")
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
type = "text",
|
||||
text = message.Content,
|
||||
cache_control = new
|
||||
{
|
||||
type = "ephemeral"
|
||||
}
|
||||
});
|
||||
}
|
||||
else
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
role = message.Role,
|
||||
content = message.Content
|
||||
});
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
if (message.Role == "system")
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
type = "text",
|
||||
text = message.Content
|
||||
});
|
||||
}
|
||||
else if (message.Role == "tool")
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
role = "user",
|
||||
content = new List<object>
|
||||
{
|
||||
new
|
||||
{
|
||||
type = "tool_result",
|
||||
tool_use_id = message.ToolCallId,
|
||||
content = message.Content
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
else if (message.Role == "assistant")
|
||||
{
|
||||
// {
|
||||
// "role": "assistant",
|
||||
// "content": [
|
||||
// {
|
||||
// "type": "text",
|
||||
// "text": "<thinking>I need to use get_weather, and the user wants SF, which is likely San Francisco, CA.</thinking>"
|
||||
// },
|
||||
// {
|
||||
// "type": "tool_use",
|
||||
// "id": "toolu_01A09q90qw90lq917835lq9",
|
||||
// "name": "get_weather",
|
||||
// "input": {
|
||||
// "location": "San Francisco, CA",
|
||||
// "unit": "celsius"
|
||||
// }
|
||||
// }
|
||||
// ]
|
||||
// },
|
||||
if (message.ToolCalls?.Count > 0)
|
||||
{
|
||||
var content = new List<object>();
|
||||
if (!string.IsNullOrEmpty(message.Content))
|
||||
{
|
||||
content.Add(new
|
||||
{
|
||||
type = "text",
|
||||
text = message.Content
|
||||
});
|
||||
}
|
||||
|
||||
foreach (var toolCall in message.ToolCalls)
|
||||
{
|
||||
content.Add(new
|
||||
{
|
||||
type = "tool_use",
|
||||
id = toolCall.Id,
|
||||
name = toolCall.Function?.Name,
|
||||
input = JsonSerializer.Deserialize<Dictionary<string, object>>(
|
||||
toolCall.Function?.Arguments, ThorJsonSerializer.DefaultOptions)
|
||||
});
|
||||
}
|
||||
|
||||
list.Add(new
|
||||
{
|
||||
role = "assistant",
|
||||
content
|
||||
});
|
||||
}
|
||||
else
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
role = "assistant",
|
||||
content = new List<object>
|
||||
{
|
||||
new
|
||||
{
|
||||
type = "text",
|
||||
text = message.Content
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
list.Add(new
|
||||
{
|
||||
role = message.Role,
|
||||
content = message.Content
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return list;
|
||||
}
|
||||
|
||||
|
||||
public async IAsyncEnumerable<ThorChatCompletionsResponse> CompleteChatStreamAsync(AiModelDescribe options,
|
||||
ThorChatCompletionsRequest input,
|
||||
CancellationToken cancellationToken)
|
||||
{
|
||||
using var openai =
|
||||
Activity.Current?.Source.StartActivity("Claudia 对话补全");
|
||||
|
||||
if (string.IsNullOrEmpty(options.Endpoint))
|
||||
{
|
||||
options.Endpoint = "https://api.anthropic.com/";
|
||||
}
|
||||
|
||||
var client = httpClientFactory.CreateClient();
|
||||
|
||||
var headers = new Dictionary<string, string>
|
||||
{
|
||||
{ "x-api-key", options.ApiKey },
|
||||
{ "anthropic-version", "2023-06-01" }
|
||||
};
|
||||
|
||||
var isThinking = input.Model.EndsWith("thinking");
|
||||
input.Model = input.Model.Replace("-thinking", string.Empty);
|
||||
var budgetTokens = 1024;
|
||||
|
||||
if (input.MaxTokens is < 2048)
|
||||
{
|
||||
input.MaxTokens = 2048;
|
||||
}
|
||||
|
||||
if (input.MaxTokens != null && input.MaxTokens / 2 < 1024)
|
||||
{
|
||||
budgetTokens = input.MaxTokens.Value / (4 * 3);
|
||||
}
|
||||
|
||||
// budgetTokens最大4096
|
||||
budgetTokens = Math.Min(budgetTokens, 4096);
|
||||
|
||||
object tool_choice;
|
||||
if (input.ToolChoice is not null && input.ToolChoice.Type == "auto")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "auto",
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else if (input.ToolChoice is not null && input.ToolChoice.Type == "any")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "any",
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else if (input.ToolChoice is not null && input.ToolChoice.Type == "tool")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "tool",
|
||||
name = input.ToolChoice.Function?.Name,
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else
|
||||
{
|
||||
tool_choice = null;
|
||||
}
|
||||
|
||||
var response = await client.HttpRequestRaw(options.Endpoint.TrimEnd('/') + "/v1/messages", new
|
||||
{
|
||||
model = input.Model,
|
||||
max_tokens = input.MaxTokens ?? 2048,
|
||||
stream = true,
|
||||
tool_choice,
|
||||
system = CreateMessage(input.Messages.Where(x => x.Role == "system").ToList(), options),
|
||||
messages = CreateMessage(input.Messages.Where(x => x.Role != "system").ToList(), options),
|
||||
top_p = isThinking ? null : input.TopP,
|
||||
thinking = isThinking
|
||||
? new
|
||||
{
|
||||
type = "enabled",
|
||||
budget_tokens = budgetTokens,
|
||||
}
|
||||
: null,
|
||||
tools = input.Tools?.Select(x => new
|
||||
{
|
||||
name = x.Function?.Name,
|
||||
description = x.Function?.Description,
|
||||
input_schema = new
|
||||
{
|
||||
type = x.Function?.Parameters?.Type,
|
||||
required = x.Function?.Parameters?.Required,
|
||||
properties = x.Function?.Parameters?.Properties?.ToDictionary(y => y.Key, y => new
|
||||
{
|
||||
description = y.Value.Description,
|
||||
type = y.Value.Type,
|
||||
@enum = y.Value.Enum
|
||||
})
|
||||
}
|
||||
}).ToArray(),
|
||||
temperature = isThinking ? null : input.Temperature
|
||||
}, string.Empty, headers);
|
||||
|
||||
openai?.SetTag("Model", input.Model);
|
||||
openai?.SetTag("Response", response.StatusCode.ToString());
|
||||
|
||||
// 大于等于400的状态码都认为是异常
|
||||
if (response.StatusCode >= HttpStatusCode.BadRequest)
|
||||
{
|
||||
var error = await response.Content.ReadAsStringAsync(cancellationToken).ConfigureAwait(false);
|
||||
logger.LogError("OpenAI对话异常 请求地址:{Address}, StatusCode: {StatusCode} Response: {Response}",
|
||||
options.Endpoint,
|
||||
response.StatusCode, error);
|
||||
|
||||
throw new Exception("OpenAI对话异常" + response.StatusCode);
|
||||
}
|
||||
|
||||
using var stream = new StreamReader(await response.Content.ReadAsStreamAsync(cancellationToken));
|
||||
|
||||
using StreamReader reader = new(await response.Content.ReadAsStreamAsync(cancellationToken));
|
||||
string? line = string.Empty;
|
||||
var first = true;
|
||||
var isThink = false;
|
||||
|
||||
string? toolId = null;
|
||||
string? toolName = null;
|
||||
while ((line = await reader.ReadLineAsync(cancellationToken).ConfigureAwait(false)) != null)
|
||||
{
|
||||
line += Environment.NewLine;
|
||||
|
||||
if (line.StartsWith('{'))
|
||||
{
|
||||
logger.LogInformation("OpenAI对话异常 , StatusCode: {StatusCode} Response: {Response}", response.StatusCode,
|
||||
line);
|
||||
|
||||
throw new Exception("OpenAI对话异常" + line);
|
||||
}
|
||||
|
||||
if (line.StartsWith(OpenAIConstant.Data))
|
||||
line = line[OpenAIConstant.Data.Length..];
|
||||
|
||||
line = line.Trim();
|
||||
|
||||
if (string.IsNullOrWhiteSpace(line)) continue;
|
||||
|
||||
if (line == OpenAIConstant.Done)
|
||||
{
|
||||
break;
|
||||
}
|
||||
|
||||
if (line.StartsWith(':'))
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
if (line.StartsWith("event: ", StringComparison.OrdinalIgnoreCase))
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
var result = JsonSerializer.Deserialize<AnthropicStreamDto>(line,
|
||||
ThorJsonSerializer.DefaultOptions);
|
||||
|
||||
if (result?.Type == "content_block_delta")
|
||||
{
|
||||
if (result.Delta.Type is "text" or "text_delta")
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
Content = result.Delta.Text,
|
||||
Role = "assistant",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Id = result?.Message?.id,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
CompletionTokens = result?.Message?.Usage?.OutputTokens,
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens,
|
||||
}
|
||||
};
|
||||
}
|
||||
else if (result.Delta.Type == "input_json_delta")
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new ThorChatChoiceResponse()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
ToolCalls =
|
||||
[
|
||||
new ThorToolCall()
|
||||
{
|
||||
Id = toolId,
|
||||
Function = new ThorChatMessageFunction()
|
||||
{
|
||||
Name = toolName,
|
||||
Arguments = result.Delta.PartialJson
|
||||
}
|
||||
}
|
||||
],
|
||||
Role = "tool",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens,
|
||||
}
|
||||
};
|
||||
}
|
||||
else
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices = new List<ThorChatChoiceResponse>()
|
||||
{
|
||||
new()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
ReasoningContent = result.Delta.Thinking,
|
||||
Role = "assistant",
|
||||
}
|
||||
}
|
||||
},
|
||||
Model = input.Model,
|
||||
Id = result?.Message?.id,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
CompletionTokens = result?.Message?.Usage?.OutputTokens,
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
continue;
|
||||
}
|
||||
|
||||
if (result?.Type == "content_block_start")
|
||||
{
|
||||
if (result?.ContentBlock?.Id is not null)
|
||||
{
|
||||
toolId = result.ContentBlock.Id;
|
||||
}
|
||||
|
||||
if (result?.ContentBlock?.Name is not null)
|
||||
{
|
||||
toolName = result.ContentBlock.Name;
|
||||
}
|
||||
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new ThorChatChoiceResponse()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
ToolCalls =
|
||||
[
|
||||
new ThorToolCall()
|
||||
{
|
||||
Id = toolId,
|
||||
Function = new ThorChatMessageFunction()
|
||||
{
|
||||
Name = toolName
|
||||
}
|
||||
}
|
||||
],
|
||||
Role = "tool",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens,
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
if (result.Type == "content_block_delta")
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new ThorChatChoiceResponse()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
ToolCallId = result?.ContentBlock?.Id,
|
||||
FunctionCall = new ThorChatMessageFunction()
|
||||
{
|
||||
Name = result?.ContentBlock?.Name,
|
||||
Arguments = result?.Delta?.PartialJson
|
||||
},
|
||||
Role = "tool",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens,
|
||||
}
|
||||
};
|
||||
continue;
|
||||
}
|
||||
|
||||
if (result.Type == "message_start")
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new ThorChatChoiceResponse()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
Content = result?.Delta?.Text,
|
||||
Role = "assistant",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
PromptTokens = result?.Message?.Usage?.InputTokens,
|
||||
}
|
||||
};
|
||||
|
||||
continue;
|
||||
}
|
||||
|
||||
if (result.Type == "message_delta")
|
||||
{
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices =
|
||||
[
|
||||
new ThorChatChoiceResponse()
|
||||
{
|
||||
Message = new ThorChatMessage()
|
||||
{
|
||||
Content = result.Delta?.Text,
|
||||
Role = "assistant",
|
||||
}
|
||||
}
|
||||
],
|
||||
Model = input.Model,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
CompletionTokens = result.Usage?.OutputTokens,
|
||||
InputTokens = result.Usage?.InputTokens
|
||||
}
|
||||
};
|
||||
|
||||
continue;
|
||||
}
|
||||
|
||||
if (result.Message == null)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
var chat = CreateResponse(result.Message);
|
||||
|
||||
var content = chat?.FirstOrDefault()?.Delta;
|
||||
|
||||
if (first && string.IsNullOrWhiteSpace(content?.Content) && string.IsNullOrEmpty(content?.ReasoningContent))
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
if (first && content.Content == OpenAIConstant.ThinkStart)
|
||||
{
|
||||
isThink = true;
|
||||
continue;
|
||||
// 需要将content的内容转换到其他字段
|
||||
}
|
||||
|
||||
if (isThink && content.Content.Contains(OpenAIConstant.ThinkEnd))
|
||||
{
|
||||
isThink = false;
|
||||
// 需要将content的内容转换到其他字段
|
||||
continue;
|
||||
}
|
||||
|
||||
if (isThink)
|
||||
{
|
||||
// 需要将content的内容转换到其他字段
|
||||
foreach (var choice in chat)
|
||||
{
|
||||
choice.Delta.ReasoningContent = choice.Delta.Content;
|
||||
choice.Delta.Content = string.Empty;
|
||||
}
|
||||
}
|
||||
|
||||
first = false;
|
||||
|
||||
yield return new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices = chat,
|
||||
Model = input.Model,
|
||||
Id = result.Message.id,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
CompletionTokens = result.Message.Usage?.OutputTokens,
|
||||
PromptTokens = result.Message.Usage?.InputTokens,
|
||||
InputTokens = result.Message.Usage?.InputTokens,
|
||||
OutputTokens = result.Message.Usage?.OutputTokens,
|
||||
TotalTokens = result.Message.Usage?.InputTokens + result.Message.Usage?.OutputTokens
|
||||
}
|
||||
};
|
||||
}
|
||||
}
|
||||
|
||||
public async Task<ThorChatCompletionsResponse> CompleteChatAsync(AiModelDescribe options,
|
||||
ThorChatCompletionsRequest input,
|
||||
CancellationToken cancellationToken)
|
||||
{
|
||||
using var openai =
|
||||
Activity.Current?.Source.StartActivity("Claudia 对话补全");
|
||||
|
||||
if (string.IsNullOrEmpty(options.Endpoint))
|
||||
{
|
||||
options.Endpoint = "https://api.anthropic.com/";
|
||||
}
|
||||
|
||||
var client = httpClientFactory.CreateClient();
|
||||
|
||||
var headers = new Dictionary<string, string>
|
||||
{
|
||||
{ "x-api-key", options.ApiKey },
|
||||
{ "anthropic-version", "2023-06-01" }
|
||||
};
|
||||
|
||||
bool isThink = input.Model.EndsWith("-thinking");
|
||||
input.Model = input.Model.Replace("-thinking", string.Empty);
|
||||
|
||||
var budgetTokens = 1024;
|
||||
if (input.MaxTokens is < 2048)
|
||||
{
|
||||
input.MaxTokens = 2048;
|
||||
}
|
||||
|
||||
if (input.MaxTokens != null && input.MaxTokens / 2 < 1024)
|
||||
{
|
||||
budgetTokens = input.MaxTokens.Value / (4 * 3);
|
||||
}
|
||||
|
||||
object tool_choice;
|
||||
if (input.ToolChoice is not null && input.ToolChoice.Type == "auto")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "auto",
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else if (input.ToolChoice is not null && input.ToolChoice.Type == "any")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "any",
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else if (input.ToolChoice is not null && input.ToolChoice.Type == "tool")
|
||||
{
|
||||
tool_choice = new
|
||||
{
|
||||
type = "tool",
|
||||
name = input.ToolChoice.Function?.Name,
|
||||
disable_parallel_tool_use = false,
|
||||
};
|
||||
}
|
||||
else
|
||||
{
|
||||
tool_choice = null;
|
||||
}
|
||||
|
||||
// budgetTokens最大4096
|
||||
budgetTokens = Math.Min(budgetTokens, 4096);
|
||||
|
||||
var response = await client.PostJsonAsync(options.Endpoint.TrimEnd('/') + "/v1/messages", new
|
||||
{
|
||||
model = input.Model,
|
||||
max_tokens = input.MaxTokens ?? 2000,
|
||||
system = CreateMessage(input.Messages.Where(x => x.Role == "system").ToList(), options),
|
||||
messages = CreateMessage(input.Messages.Where(x => x.Role != "system").ToList(), options),
|
||||
top_p = isThink ? null : input.TopP,
|
||||
tool_choice,
|
||||
thinking = isThink
|
||||
? new
|
||||
{
|
||||
type = "enabled",
|
||||
budget_tokens = budgetTokens,
|
||||
}
|
||||
: null,
|
||||
tools = input.Tools?.Select(x => new
|
||||
{
|
||||
name = x.Function?.Name,
|
||||
description = x.Function?.Description,
|
||||
input_schema = new
|
||||
{
|
||||
type = x.Function?.Parameters?.Type,
|
||||
required = x.Function?.Parameters?.Required,
|
||||
properties = x.Function?.Parameters?.Properties?.ToDictionary(y => y.Key, y => new
|
||||
{
|
||||
description = y.Value.Description,
|
||||
type = y.Value.Type,
|
||||
@enum = y.Value.Enum
|
||||
})
|
||||
}
|
||||
}).ToArray(),
|
||||
temperature = isThink ? null : input.Temperature
|
||||
}, string.Empty, headers);
|
||||
|
||||
openai?.SetTag("Model", input.Model);
|
||||
openai?.SetTag("Response", response.StatusCode.ToString());
|
||||
|
||||
// 大于等于400的状态码都认为是异常
|
||||
if (response.StatusCode >= HttpStatusCode.BadRequest)
|
||||
{
|
||||
var error = await response.Content.ReadAsStringAsync(cancellationToken).ConfigureAwait(false);
|
||||
logger.LogError("OpenAI对话异常 请求地址:{Address}, StatusCode: {StatusCode} Response: {Response}",
|
||||
options.Endpoint,
|
||||
response.StatusCode, error);
|
||||
|
||||
throw new Exception("OpenAI对话异常" + response.StatusCode.ToString());
|
||||
}
|
||||
|
||||
var value =
|
||||
await response.Content.ReadFromJsonAsync<AnthropicChatCompletionDto>(ThorJsonSerializer.DefaultOptions,
|
||||
cancellationToken: cancellationToken);
|
||||
|
||||
var thor = new ThorChatCompletionsResponse()
|
||||
{
|
||||
Choices = CreateResponse(value),
|
||||
Model = input.Model,
|
||||
Id = value.id,
|
||||
Usage = new ThorUsageResponse()
|
||||
{
|
||||
CompletionTokens = value.Usage.OutputTokens,
|
||||
PromptTokens = value.Usage.InputTokens
|
||||
}
|
||||
};
|
||||
|
||||
if (value.Usage.CacheReadInputTokens != null)
|
||||
{
|
||||
thor.Usage.PromptTokensDetails ??= new ThorUsageResponsePromptTokensDetails()
|
||||
{
|
||||
CachedTokens = value.Usage.CacheReadInputTokens.Value,
|
||||
};
|
||||
|
||||
if (value.Usage.InputTokens > 0)
|
||||
{
|
||||
thor.Usage.InputTokens = value.Usage.InputTokens;
|
||||
}
|
||||
|
||||
if (value.Usage.OutputTokens > 0)
|
||||
{
|
||||
thor.Usage.CompletionTokens = value.Usage.OutputTokens;
|
||||
thor.Usage.OutputTokens = value.Usage.OutputTokens;
|
||||
}
|
||||
}
|
||||
|
||||
thor.Usage.TotalTokens = thor.Usage.InputTokens + thor.Usage.OutputTokens;
|
||||
return thor;
|
||||
}
|
||||
}
|
||||
@@ -52,7 +52,12 @@ public class AiModelEntity : Entity<Guid>, IOrderNum, ISoftDelete
|
||||
public string? ExtraInfo { get; set; }
|
||||
|
||||
/// <summary>
|
||||
/// 模型类型
|
||||
/// 模型类型(聊天/图片等)
|
||||
/// </summary>
|
||||
public ModelTypeEnum ModelType { get; set; }
|
||||
|
||||
/// <summary>
|
||||
/// 模型Api类型,现支持同一个模型id,多种接口格式
|
||||
/// </summary>
|
||||
public ModelApiTypeEnum ModelApiType { get; set; }
|
||||
}
|
||||
@@ -12,11 +12,13 @@ using Volo.Abp.Domain.Services;
|
||||
using Yi.Framework.AiHub.Domain.AiGateWay;
|
||||
using Yi.Framework.AiHub.Domain.AiGateWay.Exceptions;
|
||||
using Yi.Framework.AiHub.Domain.Entities.Model;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Consts;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.Anthropic;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.OpenAi;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.OpenAi.Embeddings;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Dtos.OpenAi.Images;
|
||||
using Yi.Framework.AiHub.Domain.Shared.Enums;
|
||||
using Yi.Framework.Core.Extensions;
|
||||
using Yi.Framework.SqlSugarCore.Abstractions;
|
||||
using JsonSerializer = System.Text.Json.JsonSerializer;
|
||||
@@ -27,21 +29,24 @@ namespace Yi.Framework.AiHub.Domain.Managers;
|
||||
public class AiGateWayManager : DomainService
|
||||
{
|
||||
private readonly ISqlSugarRepository<AiAppAggregateRoot> _aiAppRepository;
|
||||
private readonly ISqlSugarRepository<AiModelEntity> _aiModelRepository;
|
||||
private readonly ILogger<AiGateWayManager> _logger;
|
||||
private readonly AiMessageManager _aiMessageManager;
|
||||
private readonly UsageStatisticsManager _usageStatisticsManager;
|
||||
private readonly ISpecialCompatible _specialCompatible;
|
||||
private PremiumPackageManager? _premiumPackageManager;
|
||||
|
||||
|
||||
public AiGateWayManager(ISqlSugarRepository<AiAppAggregateRoot> aiAppRepository, ILogger<AiGateWayManager> logger,
|
||||
AiMessageManager aiMessageManager, UsageStatisticsManager usageStatisticsManager,
|
||||
ISpecialCompatible specialCompatible)
|
||||
ISpecialCompatible specialCompatible, ISqlSugarRepository<AiModelEntity> aiModelRepository)
|
||||
{
|
||||
_aiAppRepository = aiAppRepository;
|
||||
_logger = logger;
|
||||
_aiMessageManager = aiMessageManager;
|
||||
_usageStatisticsManager = usageStatisticsManager;
|
||||
_specialCompatible = specialCompatible;
|
||||
_aiModelRepository = aiModelRepository;
|
||||
}
|
||||
|
||||
private PremiumPackageManager PremiumPackageManager =>
|
||||
@@ -50,17 +55,17 @@ public class AiGateWayManager : DomainService
|
||||
/// <summary>
|
||||
/// 获取模型
|
||||
/// </summary>
|
||||
/// <param name="modelApiType"></param>
|
||||
/// <param name="modelId"></param>
|
||||
/// <returns></returns>
|
||||
private async Task<AiModelDescribe> GetModelAsync(string modelId)
|
||||
private async Task<AiModelDescribe> GetModelAsync(ModelApiTypeEnum modelApiType, string modelId)
|
||||
{
|
||||
var allApp = await _aiAppRepository._DbQueryable.Includes(x => x.AiModels).ToListAsync();
|
||||
foreach (var app in allApp)
|
||||
{
|
||||
var model = app.AiModels.FirstOrDefault(x => x.ModelId == modelId);
|
||||
if (model is not null)
|
||||
{
|
||||
return new AiModelDescribe
|
||||
var aiModelDescribe = await _aiModelRepository._DbQueryable
|
||||
.LeftJoin<AiAppAggregateRoot>((model, app) => model.AiAppId == app.Id)
|
||||
.Where((model, app) => model.ModelId == modelId)
|
||||
.Where((model, app) => model.ModelApiType == modelApiType)
|
||||
.Select((model, app) =>
|
||||
new AiModelDescribe
|
||||
{
|
||||
AppId = app.Id,
|
||||
AppName = app.Name,
|
||||
@@ -73,11 +78,14 @@ public class AiGateWayManager : DomainService
|
||||
Description = model.Description,
|
||||
AppExtraUrl = app.ExtraUrl,
|
||||
ModelExtraInfo = model.ExtraInfo
|
||||
};
|
||||
}
|
||||
})
|
||||
.FirstAsync();
|
||||
if (aiModelDescribe is null)
|
||||
{
|
||||
throw new UserFriendlyException($"【{modelId}】模型当前版本【{modelApiType}】格式不支持");
|
||||
}
|
||||
|
||||
throw new UserFriendlyException($"{modelId}模型当前版本不支持");
|
||||
return aiModelDescribe;
|
||||
}
|
||||
|
||||
|
||||
@@ -92,7 +100,7 @@ public class AiGateWayManager : DomainService
|
||||
[EnumeratorCancellation] CancellationToken cancellationToken)
|
||||
{
|
||||
_specialCompatible.Compatible(request);
|
||||
var modelDescribe = await GetModelAsync(request.Model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.OpenAi, request.Model);
|
||||
var chatService =
|
||||
LazyServiceProvider.GetRequiredKeyedService<IChatCompletionService>(modelDescribe.HandlerName);
|
||||
|
||||
@@ -122,7 +130,7 @@ public class AiGateWayManager : DomainService
|
||||
var response = httpContext.Response;
|
||||
// 设置响应头,声明是 json
|
||||
//response.ContentType = "application/json; charset=UTF-8";
|
||||
var modelDescribe = await GetModelAsync(request.Model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.OpenAi, request.Model);
|
||||
var chatService =
|
||||
LazyServiceProvider.GetRequiredKeyedService<IChatCompletionService>(modelDescribe.HandlerName);
|
||||
var data = await chatService.CompleteChatAsync(modelDescribe, request, cancellationToken);
|
||||
@@ -277,6 +285,16 @@ public class AiGateWayManager : DomainService
|
||||
});
|
||||
|
||||
await _usageStatisticsManager.SetUsageAsync(userId, request.Model, tokenUsage);
|
||||
|
||||
// 扣减尊享token包用量
|
||||
if (userId is not null && PremiumPackageConst.ModeIds.Contains(request.Model))
|
||||
{
|
||||
var totalTokens = tokenUsage.TotalTokens ?? 0;
|
||||
if (totalTokens > 0)
|
||||
{
|
||||
await PremiumPackageManager.ConsumeTokensAsync(userId.Value, totalTokens);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -297,7 +315,7 @@ public class AiGateWayManager : DomainService
|
||||
var model = request.Model;
|
||||
if (string.IsNullOrEmpty(model)) model = "dall-e-2";
|
||||
|
||||
var modelDescribe = await GetModelAsync(model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.OpenAi, model);
|
||||
|
||||
// 获取渠道指定的实现类型的服务
|
||||
var imageService =
|
||||
@@ -329,6 +347,16 @@ public class AiGateWayManager : DomainService
|
||||
});
|
||||
|
||||
await _usageStatisticsManager.SetUsageAsync(userId, model, response.Usage);
|
||||
|
||||
// 扣减尊享token包用量
|
||||
if (userId is not null && PremiumPackageConst.ModeIds.Contains(request.Model))
|
||||
{
|
||||
var totalTokens = response.Usage.TotalTokens ?? 0;
|
||||
if (totalTokens > 0)
|
||||
{
|
||||
await PremiumPackageManager.ConsumeTokensAsync(userId.Value, totalTokens);
|
||||
}
|
||||
}
|
||||
}
|
||||
catch (Exception e)
|
||||
{
|
||||
@@ -357,7 +385,7 @@ public class AiGateWayManager : DomainService
|
||||
using var embedding =
|
||||
Activity.Current?.Source.StartActivity("向量模型调用");
|
||||
|
||||
var modelDescribe = await GetModelAsync(input.Model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.OpenAi, input.Model);
|
||||
|
||||
// 获取渠道指定的实现类型的服务
|
||||
var embeddingService =
|
||||
@@ -461,7 +489,7 @@ public class AiGateWayManager : DomainService
|
||||
[EnumeratorCancellation] CancellationToken cancellationToken)
|
||||
{
|
||||
_specialCompatible.AnthropicCompatible(request);
|
||||
var modelDescribe = await GetModelAsync(request.Model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.Claude, request.Model);
|
||||
var chatService =
|
||||
LazyServiceProvider.GetRequiredKeyedService<IAnthropicChatCompletionService>(modelDescribe.HandlerName);
|
||||
|
||||
@@ -491,7 +519,7 @@ public class AiGateWayManager : DomainService
|
||||
var response = httpContext.Response;
|
||||
// 设置响应头,声明是 json
|
||||
//response.ContentType = "application/json; charset=UTF-8";
|
||||
var modelDescribe = await GetModelAsync(request.Model);
|
||||
var modelDescribe = await GetModelAsync(ModelApiTypeEnum.Claude, request.Model);
|
||||
var chatService =
|
||||
LazyServiceProvider.GetRequiredKeyedService<IAnthropicChatCompletionService>(modelDescribe.HandlerName);
|
||||
var data = await chatService.ChatCompletionsAsync(modelDescribe, request, cancellationToken);
|
||||
@@ -516,14 +544,10 @@ public class AiGateWayManager : DomainService
|
||||
await _usageStatisticsManager.SetUsageAsync(userId.Value, request.Model, data.TokenUsage);
|
||||
|
||||
// 扣减尊享token包用量
|
||||
var totalTokens = data.TokenUsage.TotalTokens??0;
|
||||
var totalTokens = data.TokenUsage.TotalTokens ?? 0;
|
||||
if (totalTokens > 0)
|
||||
{
|
||||
var consumeSuccess = await PremiumPackageManager.ConsumeTokensAsync(userId.Value, totalTokens);
|
||||
if (!consumeSuccess)
|
||||
{
|
||||
_logger.LogWarning($"用户 {userId.Value} 尊享token包扣减失败,消耗token数: {totalTokens}");
|
||||
}
|
||||
await PremiumPackageManager.ConsumeTokensAsync(userId.Value, totalTokens);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -562,10 +586,11 @@ public class AiGateWayManager : DomainService
|
||||
await foreach (var responseResult in completeChatResponse)
|
||||
{
|
||||
//message_start是为了保底机制
|
||||
if (responseResult.Item1.Contains("message_delta")||responseResult.Item1.Contains("message_start"))
|
||||
if (responseResult.Item1.Contains("message_delta") || responseResult.Item1.Contains("message_start"))
|
||||
{
|
||||
tokenUsage = responseResult.Item2?.TokenUsage;
|
||||
}
|
||||
|
||||
backupSystemContent.Append(responseResult.Item2?.Delta?.Text);
|
||||
await WriteAsEventStreamDataAsync(httpContext, responseResult.Item1, responseResult.Item2,
|
||||
cancellationToken);
|
||||
@@ -622,7 +647,7 @@ public class AiGateWayManager : DomainService
|
||||
// 扣减尊享token包用量
|
||||
if (userId.HasValue && tokenUsage is not null)
|
||||
{
|
||||
var totalTokens = tokenUsage.TotalTokens??0;
|
||||
var totalTokens = tokenUsage.TotalTokens ?? 0;
|
||||
if (totalTokens > 0)
|
||||
{
|
||||
var consumeSuccess = await PremiumPackageManager.ConsumeTokensAsync(userId.Value, totalTokens);
|
||||
|
||||
@@ -42,7 +42,8 @@ namespace Yi.Framework.AiHub.Domain
|
||||
nameof(DeepSeekChatCompletionsService));
|
||||
services.AddKeyedTransient<IChatCompletionService, OpenAiChatCompletionsService>(
|
||||
nameof(OpenAiChatCompletionsService));
|
||||
|
||||
services.AddKeyedTransient<IChatCompletionService, ClaudiaChatCompletionsService>(
|
||||
nameof(ClaudiaChatCompletionsService));
|
||||
#endregion
|
||||
|
||||
#region Anthropic ChatCompletion
|
||||
|
||||
Reference in New Issue
Block a user