Merge branch 'songquanpeng:main' into base

2023-12-18 23:19:28 +08:00 · 2023-12-18 23:19:28 +08:00 · 52ac1fe7d8
commit 52ac1fe7d8
parent cf0b2c8858 97030e27f8
17 changed files with 479 additions and 46 deletions
--- a/README.en.md
+++ b/README.en.md
@ -60,7 +60,7 @@ _✨ Access all LLM through the standard OpenAI API format, easy to deploy & use
 1. Support for multiple large models:
   + [x] [OpenAI ChatGPT Series Models](https://platform.openai.com/docs/guides/gpt/chat-completions-api) (Supports [Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference))
   + [x] [Anthropic Claude Series Models](https://anthropic.com)
-   + [x] [Google PaLM2 Series Models](https://developers.generativeai.google)
+   + [x] [Google PaLM2 and Gemini Series Models](https://developers.generativeai.google)
   + [x] [Baidu Wenxin Yiyuan Series Models](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
   + [x] [Alibaba Tongyi Qianwen Series Models](https://help.aliyun.com/document_detail/2400395.html)
   + [x] [Zhipu ChatGLM Series Models](https://bigmodel.cn)
--- a/README.ja.md
+++ b/README.ja.md
@ -60,7 +60,7 @@ _✨ 標準的な OpenAI API フォーマットを通じてすべての LLM に
 1. 複数の大型モデルをサポート:
   + [x] [OpenAI ChatGPT シリーズモデル](https://platform.openai.com/docs/guides/gpt/chat-completions-api) ([Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference) をサポート)
   + [x] [Anthropic Claude シリーズモデル](https://anthropic.com)
-   + [x] [Google PaLM2 シリーズモデル](https://developers.generativeai.google)
+   + [x] [Google PaLM2/Gemini シリーズモデル](https://developers.generativeai.google)
   + [x] [Baidu Wenxin Yiyuan シリーズモデル](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
   + [x] [Alibaba Tongyi Qianwen シリーズモデル](https://help.aliyun.com/document_detail/2400395.html)
   + [x] [Zhipu ChatGLM シリーズモデル](https://bigmodel.cn)
--- a/README.md
+++ b/README.md
@ -66,20 +66,14 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
 1. 支持多种大模型：
   + [x] [OpenAI ChatGPT 系列模型](https://platform.openai.com/docs/guides/gpt/chat-completions-api)（支持 [Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference)）
   + [x] [Anthropic Claude 系列模型](https://anthropic.com)
-   + [x] [Google PaLM2 系列模型](https://developers.generativeai.google)
+   + [x] [Google PaLM2/Gemini 系列模型](https://developers.generativeai.google)
   + [x] [百度文心一言系列模型](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
   + [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
   + [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)
   + [x] [智谱 ChatGLM 系列模型](https://bigmodel.cn)
   + [x] [360 智脑](https://ai.360.cn)
   + [x] [腾讯混元大模型](https://cloud.tencent.com/document/product/1729)
-2. 支持配置镜像以及众多第三方代理服务：
-   + [x] [OpenAI-SB](https://openai-sb.com)
-   + [x] [CloseAI](https://referer.shadowai.xyz/r/2412)
-   + [x] [API2D](https://api2d.com/r/197971)
-   + [x] [OhMyGPT](https://aigptx.top?aff=uFpUl2Kf)
-   + [x] [AI Proxy](https://aiproxy.io/?i=OneAPI) （邀请码：`OneAPI`）
-   + [x] 自定义渠道：例如各种未收录的第三方代理服务
+2. 支持配置镜像以及众多[第三方代理服务](https://iamazing.cn/page/openai-api-third-party-services)。
 3. 支持通过**负载均衡**的方式访问多个渠道。
 4. 支持 **stream 模式**，可以通过流式传输实现打字机效果。
 5. 支持**多机部署**，[详见此处](#多机部署)。
--- a/common/constants.go
+++ b/common/constants.go
@ -187,6 +187,7 @@ const (
 	ChannelTypeAIProxyLibrary = 21
 	ChannelTypeFastGPT        = 22
 	ChannelTypeTencent        = 23
+	ChannelTypeGemini         = 24
 )

 var ChannelBaseURLs = []string{
@ -214,4 +215,5 @@ var ChannelBaseURLs = []string{
 	"https://api.aiproxy.io",            // 21
 	"https://fastgpt.run/api/openapi",   // 22
 	"https://hunyuan.cloud.tencent.com", //23
+	"",                                  //24
 }
--- a/common/model-ratio.go
+++ b/common/model-ratio.go
@ -83,12 +83,15 @@ var ModelRatio = map[string]float64{
 	"ERNIE-Bot-4":               8.572,  // ￥0.12 / 1k tokens
 	"Embedding-V1":              0.1429, // ￥0.002 / 1k tokens
 	"PaLM-2":                    1,
+	"gemini-pro":                1,      // $0.00025 / 1k characters -> $0.001 / 1k tokens
 	"chatglm_turbo":             0.3572, // ￥0.005 / 1k tokens
 	"chatglm_pro":               0.7143, // ￥0.01 / 1k tokens
 	"chatglm_std":               0.3572, // ￥0.005 / 1k tokens
 	"chatglm_lite":              0.1429, // ￥0.002 / 1k tokens
-	"qwen-turbo":                0.8572, // ￥0.012 / 1k tokens
-	"qwen-plus":                 10,     // ￥0.14 / 1k tokens
+	"qwen-turbo":                0.5715, // ￥0.008 / 1k tokens  // https://help.aliyun.com/zh/dashscope/developer-reference/tongyi-thousand-questions-metering-and-billing
+	"qwen-plus":                 1.4286, // ￥0.02 / 1k tokens
+	"qwen-max":                  1.4286, // ￥0.02 / 1k tokens
+	"qwen-max-longcontext":      1.4286, // ￥0.02 / 1k tokens
 	"text-embedding-v1":         0.05,   // ￥0.0007 / 1k tokens
 	"SparkDesk":                 1.2858, // ￥0.018 / 1k tokens
 	"360GPT_S2_V9":              0.8572, // ¥0.012 / 1k tokens
--- a/controller/channel-test.go
+++ b/controller/channel-test.go
@ -20,6 +20,8 @@ func testChannel(channel *model.Channel, request ChatRequest) (err error, openai
 	switch channel.Type {
 	case common.ChannelTypePaLM:
 		fallthrough
+	case common.ChannelTypeGemini:
+		fallthrough
 	case common.ChannelTypeAnthropic:
 		fallthrough
 	case common.ChannelTypeBaidu:
--- a/controller/model.go
+++ b/controller/model.go
@ -423,6 +423,15 @@ func init() {
 			Root:       "PaLM-2",
 			Parent:     nil,
 		},
+		{
+			Id:         "gemini-pro",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "google",
+			Permission: permission,
+			Root:       "gemini-pro",
+			Parent:     nil,
+		},
 		{
 			Id:         "chatglm_turbo",
 			Object:     "model",
@ -477,6 +486,24 @@ func init() {
 			Root:       "qwen-plus",
 			Parent:     nil,
 		},
+		{
+			Id:         "qwen-max",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "ali",
+			Permission: permission,
+			Root:       "qwen-max",
+			Parent:     nil,
+		},
+		{
+			Id:         "qwen-max-longcontext",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "ali",
+			Permission: permission,
+			Root:       "qwen-max-longcontext",
+			Parent:     nil,
+		},
 		{
 			Id:         "text-embedding-v1",
 			Object:     "model",
--- a/controller/relay-ali.go
+++ b/controller/relay-ali.go
@ -13,13 +13,13 @@ import (
 // https://help.aliyun.com/document_detail/613695.html?spm=a2c4g.2399480.0.0.1adb778fAdzP9w#341800c0f8w0r

 type AliMessage struct {
-	User string `json:"user"`
-	Bot  string `json:"bot"`
+	Content string `json:"content"`
+	Role    string `json:"role"`
 }

 type AliInput struct {
-	Prompt  string       `json:"prompt"`
-	History []AliMessage `json:"history"`
+	//Prompt   string       `json:"prompt"`
+	Messages []AliMessage `json:"messages"`
 }

 type AliParameters struct {
@ -83,32 +83,17 @@ type AliChatResponse struct {

 func requestOpenAI2Ali(request GeneralOpenAIRequest) *AliChatRequest {
 	messages := make([]AliMessage, 0, len(request.Messages))
-	prompt := ""
 	for i := 0; i < len(request.Messages); i++ {
 		message := request.Messages[i]
-		if message.Role == "system" {
 		messages = append(messages, AliMessage{
-				User: message.StringContent(),
-				Bot:  "Okay",
+			Content: message.StringContent(),
+			Role:    strings.ToLower(message.Role),
 		})
-			continue
-		} else {
-			if i == len(request.Messages)-1 {
-				prompt = message.StringContent()
-				break
-			}
-			messages = append(messages, AliMessage{
-				User: message.StringContent(),
-				Bot:  request.Messages[i+1].StringContent(),
-			})
-			i++
-		}
 	}
 	return &AliChatRequest{
 		Model: request.Model,
 		Input: AliInput{
-			Prompt:  prompt,
-			History: messages,
+			Messages: messages,
 		},
 		//Parameters: AliParameters{  // ChatGPT's parameters are not compatible with Ali's
 		//	TopP: request.TopP,
--- a/controller/relay-gemini.go
+++ b/controller/relay-gemini.go
@ -0,0 +1,305 @@
+package controller
+
+import (
+	"bufio"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"one-api/common"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+)
+
+type GeminiChatRequest struct {
+	Contents         []GeminiChatContent        `json:"contents"`
+	SafetySettings   []GeminiChatSafetySettings `json:"safety_settings,omitempty"`
+	GenerationConfig GeminiChatGenerationConfig `json:"generation_config,omitempty"`
+	Tools            []GeminiChatTools          `json:"tools,omitempty"`
+}
+
+type GeminiInlineData struct {
+	MimeType string `json:"mimeType"`
+	Data     string `json:"data"`
+}
+
+type GeminiPart struct {
+	Text       string            `json:"text,omitempty"`
+	InlineData *GeminiInlineData `json:"inlineData,omitempty"`
+}
+
+type GeminiChatContent struct {
+	Role  string       `json:"role,omitempty"`
+	Parts []GeminiPart `json:"parts"`
+}
+
+type GeminiChatSafetySettings struct {
+	Category  string `json:"category"`
+	Threshold string `json:"threshold"`
+}
+
+type GeminiChatTools struct {
+	FunctionDeclarations any `json:"functionDeclarations,omitempty"`
+}
+
+type GeminiChatGenerationConfig struct {
+	Temperature     float64  `json:"temperature,omitempty"`
+	TopP            float64  `json:"topP,omitempty"`
+	TopK            float64  `json:"topK,omitempty"`
+	MaxOutputTokens int      `json:"maxOutputTokens,omitempty"`
+	CandidateCount  int      `json:"candidateCount,omitempty"`
+	StopSequences   []string `json:"stopSequences,omitempty"`
+}
+
+// Setting safety to the lowest possible values since Gemini is already powerless enough
+func requestOpenAI2Gemini(textRequest GeneralOpenAIRequest) *GeminiChatRequest {
+	geminiRequest := GeminiChatRequest{
+		Contents: make([]GeminiChatContent, 0, len(textRequest.Messages)),
+		//SafetySettings: []GeminiChatSafetySettings{
+		//	{
+		//		Category:  "HARM_CATEGORY_HARASSMENT",
+		//		Threshold: "BLOCK_ONLY_HIGH",
+		//	},
+		//	{
+		//		Category:  "HARM_CATEGORY_HATE_SPEECH",
+		//		Threshold: "BLOCK_ONLY_HIGH",
+		//	},
+		//	{
+		//		Category:  "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+		//		Threshold: "BLOCK_ONLY_HIGH",
+		//	},
+		//	{
+		//		Category:  "HARM_CATEGORY_DANGEROUS_CONTENT",
+		//		Threshold: "BLOCK_ONLY_HIGH",
+		//	},
+		//},
+		GenerationConfig: GeminiChatGenerationConfig{
+			Temperature:     textRequest.Temperature,
+			TopP:            textRequest.TopP,
+			MaxOutputTokens: textRequest.MaxTokens,
+		},
+	}
+	if textRequest.Functions != nil {
+		geminiRequest.Tools = []GeminiChatTools{
+			{
+				FunctionDeclarations: textRequest.Functions,
+			},
+		}
+	}
+	shouldAddDummyModelMessage := false
+	for _, message := range textRequest.Messages {
+		content := GeminiChatContent{
+			Role: message.Role,
+			Parts: []GeminiPart{
+				{
+					Text: message.StringContent(),
+				},
+			},
+		}
+		// there's no assistant role in gemini and API shall vomit if Role is not user or model
+		if content.Role == "assistant" {
+			content.Role = "model"
+		}
+		// Converting system prompt to prompt from user for the same reason
+		if content.Role == "system" {
+			content.Role = "user"
+			shouldAddDummyModelMessage = true
+		}
+		geminiRequest.Contents = append(geminiRequest.Contents, content)
+
+		// If a system message is the last message, we need to add a dummy model message to make gemini happy
+		if shouldAddDummyModelMessage {
+			geminiRequest.Contents = append(geminiRequest.Contents, GeminiChatContent{
+				Role: "model",
+				Parts: []GeminiPart{
+					{
+						Text: "Okay",
+					},
+				},
+			})
+			shouldAddDummyModelMessage = false
+		}
+	}
+
+	return &geminiRequest
+}
+
+type GeminiChatResponse struct {
+	Candidates     []GeminiChatCandidate    `json:"candidates"`
+	PromptFeedback GeminiChatPromptFeedback `json:"promptFeedback"`
+}
+
+func (g *GeminiChatResponse) GetResponseText() string {
+	if g == nil {
+		return ""
+	}
+	if len(g.Candidates) > 0 && len(g.Candidates[0].Content.Parts) > 0 {
+		return g.Candidates[0].Content.Parts[0].Text
+	}
+	return ""
+}
+
+type GeminiChatCandidate struct {
+	Content       GeminiChatContent        `json:"content"`
+	FinishReason  string                   `json:"finishReason"`
+	Index         int64                    `json:"index"`
+	SafetyRatings []GeminiChatSafetyRating `json:"safetyRatings"`
+}
+
+type GeminiChatSafetyRating struct {
+	Category    string `json:"category"`
+	Probability string `json:"probability"`
+}
+
+type GeminiChatPromptFeedback struct {
+	SafetyRatings []GeminiChatSafetyRating `json:"safetyRatings"`
+}
+
+func responseGeminiChat2OpenAI(response *GeminiChatResponse) *OpenAITextResponse {
+	fullTextResponse := OpenAITextResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", common.GetUUID()),
+		Object:  "chat.completion",
+		Created: common.GetTimestamp(),
+		Choices: make([]OpenAITextResponseChoice, 0, len(response.Candidates)),
+	}
+	for i, candidate := range response.Candidates {
+		choice := OpenAITextResponseChoice{
+			Index: i,
+			Message: Message{
+				Role:    "assistant",
+				Content: "",
+			},
+			FinishReason: stopFinishReason,
+		}
+		if len(candidate.Content.Parts) > 0 {
+			choice.Message.Content = candidate.Content.Parts[0].Text
+		}
+		fullTextResponse.Choices = append(fullTextResponse.Choices, choice)
+	}
+	return &fullTextResponse
+}
+
+func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) *ChatCompletionsStreamResponse {
+	var choice ChatCompletionsStreamResponseChoice
+	choice.Delta.Content = geminiResponse.GetResponseText()
+	choice.FinishReason = &stopFinishReason
+	var response ChatCompletionsStreamResponse
+	response.Object = "chat.completion.chunk"
+	response.Model = "gemini"
+	response.Choices = []ChatCompletionsStreamResponseChoice{choice}
+	return &response
+}
+
+func geminiChatStreamHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithStatusCode, string) {
+	responseText := ""
+	dataChan := make(chan string)
+	stopChan := make(chan bool)
+	scanner := bufio.NewScanner(resp.Body)
+	scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
+		if atEOF && len(data) == 0 {
+			return 0, nil, nil
+		}
+		if i := strings.Index(string(data), "\n"); i >= 0 {
+			return i + 1, data[0:i], nil
+		}
+		if atEOF {
+			return len(data), data, nil
+		}
+		return 0, nil, nil
+	})
+	go func() {
+		for scanner.Scan() {
+			data := scanner.Text()
+			data = strings.TrimSpace(data)
+			if !strings.HasPrefix(data, "\"text\": \"") {
+				continue
+			}
+			data = strings.TrimPrefix(data, "\"text\": \"")
+			data = strings.TrimSuffix(data, "\"")
+			dataChan <- data
+		}
+		stopChan <- true
+	}()
+	setEventStreamHeaders(c)
+	c.Stream(func(w io.Writer) bool {
+		select {
+		case data := <-dataChan:
+			// this is used to prevent annoying \ related format bug
+			data = fmt.Sprintf("{\"content\": \"%s\"}", data)
+			type dummyStruct struct {
+				Content string `json:"content"`
+			}
+			var dummy dummyStruct
+			err := json.Unmarshal([]byte(data), &dummy)
+			responseText += dummy.Content
+			var choice ChatCompletionsStreamResponseChoice
+			choice.Delta.Content = dummy.Content
+			response := ChatCompletionsStreamResponse{
+				Id:      fmt.Sprintf("chatcmpl-%s", common.GetUUID()),
+				Object:  "chat.completion.chunk",
+				Created: common.GetTimestamp(),
+				Model:   "gemini-pro",
+				Choices: []ChatCompletionsStreamResponseChoice{choice},
+			}
+			jsonResponse, err := json.Marshal(response)
+			if err != nil {
+				common.SysError("error marshalling stream response: " + err.Error())
+				return true
+			}
+			c.Render(-1, common.CustomEvent{Data: "data: " + string(jsonResponse)})
+			return true
+		case <-stopChan:
+			c.Render(-1, common.CustomEvent{Data: "data: [DONE]"})
+			return false
+		}
+	})
+	err := resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), ""
+	}
+	return nil, responseText
+}
+
+func geminiChatHandler(c *gin.Context, resp *http.Response, promptTokens int, model string) (*OpenAIErrorWithStatusCode, *Usage) {
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return errorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+	var geminiResponse GeminiChatResponse
+	err = json.Unmarshal(responseBody, &geminiResponse)
+	if err != nil {
+		return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	if len(geminiResponse.Candidates) == 0 {
+		return &OpenAIErrorWithStatusCode{
+			OpenAIError: OpenAIError{
+				Message: "No candidates returned",
+				Type:    "server_error",
+				Param:   "",
+				Code:    500,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+	fullTextResponse := responseGeminiChat2OpenAI(&geminiResponse)
+	completionTokens := countTokenText(geminiResponse.GetResponseText(), model)
+	usage := Usage{
+		PromptTokens:     promptTokens,
+		CompletionTokens: completionTokens,
+		TotalTokens:      promptTokens + completionTokens,
+	}
+	fullTextResponse.Usage = usage
+	jsonResponse, err := json.Marshal(fullTextResponse)
+	if err != nil {
+		return errorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(jsonResponse)
+	return nil, &usage
+}
--- a/controller/relay-image.go
+++ b/controller/relay-image.go
@ -19,7 +19,6 @@ func isWithinRange(element string, value int) bool {
 	if _, ok := common.DalleGenerationImageAmounts[element]; !ok {
 		return false
 	}
-
 	min := common.DalleGenerationImageAmounts[element][0]
 	max := common.DalleGenerationImageAmounts[element][1]

@ -42,6 +41,10 @@ func relayImageHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 		return errorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
 	}

+	if imageRequest.N == 0 {
+		imageRequest.N = 1
+	}
+
 	// Size validation
 	if imageRequest.Size != "" {
 		imageSize = imageRequest.Size
@ -79,8 +82,11 @@ func relayImageHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode

 	// Number of generated images validation
 	if isWithinRange(imageModel, imageRequest.N) == false {
+		// channel not azure
+		if channelType != common.ChannelTypeAzure {
 			return errorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
 		}
+	}

 	// map model name
 	modelMapping := c.GetString("model_mapping")
@ -102,7 +108,7 @@ func relayImageHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 		baseURL = c.GetString("base_url")
 	}
 	fullRequestURL := getFullRequestURL(baseURL, requestURL, channelType)
-	if channelType == common.ChannelTypeAzure && relayMode == RelayModeImagesGenerations {
+	if channelType == common.ChannelTypeAzure {
 		// https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart?tabs=dalle3%2Ccommand-line&pivots=rest-api
 		apiVersion := GetAPIVersion(c)
 		// https://{resource_name}.openai.azure.com/openai/deployments/dall-e-3/images/generations?api-version=2023-06-01-preview
--- a/controller/relay-text.go
+++ b/controller/relay-text.go
@ -27,6 +27,7 @@ const (
 	APITypeXunfei
 	APITypeAIProxyLibrary
 	APITypeTencent
+	APITypeGemini
 )

 var httpClient *http.Client
@ -118,6 +119,8 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		apiType = APITypeAIProxyLibrary
 	case common.ChannelTypeTencent:
 		apiType = APITypeTencent
+	case common.ChannelTypeGemini:
+		apiType = APITypeGemini
 	}
 	baseURL := common.ChannelBaseURLs[channelType]
 	requestURL := c.Request.URL.String()
@ -177,6 +180,23 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		apiKey := c.Request.Header.Get("Authorization")
 		apiKey = strings.TrimPrefix(apiKey, "Bearer ")
 		fullRequestURL += "?key=" + apiKey
+	case APITypeGemini:
+		requestBaseURL := "https://generativelanguage.googleapis.com"
+		if baseURL != "" {
+			requestBaseURL = baseURL
+		}
+		version := "v1"
+		if c.GetString("api_version") != "" {
+			version = c.GetString("api_version")
+		}
+		action := "generateContent"
+		if textRequest.Stream {
+			action = "streamGenerateContent"
+		}
+		fullRequestURL = fmt.Sprintf("%s/%s/models/%s:%s", requestBaseURL, version, textRequest.Model, action)
+		apiKey := c.Request.Header.Get("Authorization")
+		apiKey = strings.TrimPrefix(apiKey, "Bearer ")
+		fullRequestURL += "?key=" + apiKey
 	case APITypeZhipu:
 		method := "invoke"
 		if textRequest.Stream {
@ -274,6 +294,13 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			return errorWrapper(err, "marshal_text_request_failed", http.StatusInternalServerError)
 		}
 		requestBody = bytes.NewBuffer(jsonStr)
+	case APITypeGemini:
+		geminiChatRequest := requestOpenAI2Gemini(textRequest)
+		jsonStr, err := json.Marshal(geminiChatRequest)
+		if err != nil {
+			return errorWrapper(err, "marshal_text_request_failed", http.StatusInternalServerError)
+		}
+		requestBody = bytes.NewBuffer(jsonStr)
 	case APITypeZhipu:
 		zhipuRequest := requestOpenAI2Zhipu(textRequest)
 		jsonStr, err := json.Marshal(zhipuRequest)
@ -367,6 +394,8 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			req.Header.Set("Authorization", apiKey)
 		case APITypePaLM:
 			// do not set Authorization header
+		case APITypeGemini:
+			// do not set Authorization header
 		default:
 			req.Header.Set("Authorization", "Bearer "+apiKey)
 		}
@ -527,6 +556,25 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			}
 			return nil
 		}
+	case APITypeGemini:
+		if textRequest.Stream {
+			err, responseText := geminiChatStreamHandler(c, resp)
+			if err != nil {
+				return err
+			}
+			textResponse.Usage.PromptTokens = promptTokens
+			textResponse.Usage.CompletionTokens = countTokenText(responseText, textRequest.Model)
+			return nil
+		} else {
+			err, usage := geminiChatHandler(c, resp, promptTokens, textRequest.Model)
+			if err != nil {
+				return err
+			}
+			if usage != nil {
+				textResponse.Usage = *usage
+			}
+			return nil
+		}
 	case APITypeZhipu:
 		if isStream {
 			err, usage := zhipuStreamHandler(c, resp)
--- a/controller/relay-utils.go
+++ b/controller/relay-utils.go
@ -263,11 +263,52 @@ func setEventStreamHeaders(c *gin.Context) {
 	c.Writer.Header().Set("X-Accel-Buffering", "no")
 }

+type GeneralErrorResponse struct {
+	Error    OpenAIError `json:"error"`
+	Message  string      `json:"message"`
+	Msg      string      `json:"msg"`
+	Err      string      `json:"err"`
+	ErrorMsg string      `json:"error_msg"`
+	Header   struct {
+		Message string `json:"message"`
+	} `json:"header"`
+	Response struct {
+		Error struct {
+			Message string `json:"message"`
+		} `json:"error"`
+	} `json:"response"`
+}
+
+func (e GeneralErrorResponse) ToMessage() string {
+	if e.Error.Message != "" {
+		return e.Error.Message
+	}
+	if e.Message != "" {
+		return e.Message
+	}
+	if e.Msg != "" {
+		return e.Msg
+	}
+	if e.Err != "" {
+		return e.Err
+	}
+	if e.ErrorMsg != "" {
+		return e.ErrorMsg
+	}
+	if e.Header.Message != "" {
+		return e.Header.Message
+	}
+	if e.Response.Error.Message != "" {
+		return e.Response.Error.Message
+	}
+	return ""
+}
+
 func relayErrorHandler(resp *http.Response) (openAIErrorWithStatusCode *OpenAIErrorWithStatusCode) {
 	openAIErrorWithStatusCode = &OpenAIErrorWithStatusCode{
 		StatusCode: resp.StatusCode,
 		OpenAIError: OpenAIError{
-			Message: fmt.Sprintf("bad response status code %d", resp.StatusCode),
+			Message: "",
 			Type:    "upstream_error",
 			Code:    "bad_response_status_code",
 			Param:   strconv.Itoa(resp.StatusCode),
@ -281,12 +322,20 @@ func relayErrorHandler(resp *http.Response) (openAIErrorWithStatusCode *OpenAIEr
 	if err != nil {
 		return
 	}
-	var textResponse TextResponse
-	err = json.Unmarshal(responseBody, &textResponse)
+	var errResponse GeneralErrorResponse
+	err = json.Unmarshal(responseBody, &errResponse)
 	if err != nil {
 		return
 	}
-	openAIErrorWithStatusCode.OpenAIError = textResponse.Error
+	if errResponse.Error.Message != "" {
+		// OpenAI format error, so we override the default one
+		openAIErrorWithStatusCode.OpenAIError = errResponse.Error
+	} else {
+		openAIErrorWithStatusCode.OpenAIError.Message = errResponse.ToMessage()
+	}
+	if openAIErrorWithStatusCode.OpenAIError.Message == "" {
+		openAIErrorWithStatusCode.OpenAIError.Message = fmt.Sprintf("bad response status code %d", resp.StatusCode)
+	}
 	return
 }

--- a/controller/relay-xunfei.go
+++ b/controller/relay-xunfei.go
@ -230,7 +230,13 @@ func xunfeiHandler(c *gin.Context, textRequest GeneralOpenAIRequest, appId strin
 		case stop = <-stopChan:
 		}
 	}
-
+	if len(xunfeiResponse.Payload.Choices.Text) == 0 {
+		xunfeiResponse.Payload.Choices.Text = []XunfeiChatResponseTextItem{
+			{
+				Content: "",
+			},
+		}
+	}
 	xunfeiResponse.Payload.Choices.Text[0].Content = content

 	response := responseXunfei2OpenAI(&xunfeiResponse)
--- a/controller/relay.go
+++ b/controller/relay.go
@ -236,7 +236,7 @@ type ChatCompletionsStreamResponseChoice struct {
 	Delta struct {
 		Content string `json:"content"`
 	} `json:"delta"`
-	FinishReason *string `json:"finish_reason"`
+	FinishReason *string `json:"finish_reason,omitempty"`
 }

 type ChatCompletionsStreamResponse struct {
--- a/middleware/distributor.go
+++ b/middleware/distributor.go
@ -87,6 +87,8 @@ func Distribute() func(c *gin.Context) {
 			c.Set("api_version", channel.Other)
 		case common.ChannelTypeXunfei:
 			c.Set("api_version", channel.Other)
+		case common.ChannelTypeGemini:
+			c.Set("api_version", channel.Other)
 		case common.ChannelTypeAIProxyLibrary:
 			c.Set("library_id", channel.Other)
 		case common.ChannelTypeAli:
--- a/web/src/constants/channel.constants.js
+++ b/web/src/constants/channel.constants.js
@ -3,6 +3,7 @@ export const CHANNEL_OPTIONS = [
  { key: 14, text: 'Anthropic Claude', value: 14, color: 'black' },
  { key: 3, text: 'Azure OpenAI', value: 3, color: 'olive' },
  { key: 11, text: 'Google PaLM2', value: 11, color: 'orange' },
+  { key: 24, text: 'Google Gemini', value: 24, color: 'orange' },
  { key: 15, text: '百度文心千帆', value: 15, color: 'blue' },
  { key: 17, text: '阿里通义千问', value: 17, color: 'orange' },
  { key: 18, text: '讯飞星火认知', value: 18, color: 'blue' },
--- a/web/src/pages/Channel/EditChannel.js
+++ b/web/src/pages/Channel/EditChannel.js
@ -69,7 +69,7 @@ const EditChannel = () => {
          localModels = ['ERNIE-Bot', 'ERNIE-Bot-turbo', 'ERNIE-Bot-4', 'Embedding-V1'];
          break;
        case 17:
-          localModels = ['qwen-turbo', 'qwen-plus', 'text-embedding-v1'];
+          localModels = ['qwen-turbo', 'qwen-plus', 'qwen-max', 'qwen-max-longcontext', 'text-embedding-v1'];
          break;
        case 16:
          localModels = ['chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
@ -83,6 +83,9 @@ const EditChannel = () => {
        case 23:
          localModels = ['hunyuan'];
          break;
+        case 24:
+          localModels = ['gemini-pro'];
+          break;
      }
      setInputs((inputs) => ({ ...inputs, models: localModels }));
    }