fix: fix max_tokens check

- Add a condition to validate the n value only for non-Azure channels, ensuring it falls within the acceptable range.
- Fix Azure compatibility

README.md

@@ -73,13 +73,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [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)。
@@ -371,6 +365,7 @@ graph LR
     + `TIKTOKEN_CACHE_DIR`：默认程序启动时会联网下载一些通用的词元的编码，如：`gpt-3.5-turbo`，在一些网络环境不稳定，或者离线情况，可能会导致启动有问题，可以配置此目录缓存数据，可迁移到离线环境。
     + `DATA_GYM_CACHE_DIR`：目前该配置作用与 `TIKTOKEN_CACHE_DIR` 一致，但是优先级没有它高。
 15. `RELAY_TIMEOUT`：中继超时设置，单位为秒，默认不设置超时时间。
+16. `SQLITE_BUSY_TIMEOUT`：SQLite 锁等待超时设置，单位为毫秒，默认 `3000`。
 
 ### 命令行参数
 1. `--port <port_number>`: 指定服务器监听的端口号，默认为 `3000`。

common/database.go

@@ -4,3 +4,4 @@ var UsingSQLite = false
 var UsingPostgreSQL = false
 
 var SQLitePath = "one-api.db"
+var SQLiteBusyTimeout = GetOrDefault("SQLITE_BUSY_TIMEOUT", 3000)

common/model-ratio.go

@@ -88,8 +88,10 @@ var ModelRatio = map[string]float64{
 	"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

controller/model.go

@@ -486,6 +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",

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",
-			})
-			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++
-		}
+		messages = append(messages, AliMessage{
+			Content: message.StringContent(),
+			Role:    strings.ToLower(message.Role),
+		})
 	}
 	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,

controller/relay-gemini.go

@@ -1,11 +1,13 @@
 package controller
 
 import (
+	"bufio"
 	"encoding/json"
 	"fmt"
 	"io"
 	"net/http"
 	"one-api/common"
+	"strings"
 
 	"github.com/gin-gonic/gin"
 )
@@ -112,7 +114,7 @@ func requestOpenAI2Gemini(textRequest GeneralOpenAIRequest) *GeminiChatRequest {
 				Role: "model",
 				Parts: []GeminiPart{
 					{
-						Text: "ok",
+						Text: "Okay",
 					},
 				},
 			})
@@ -128,6 +130,16 @@ type GeminiChatResponse struct {
 	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"`
@@ -156,10 +168,13 @@ func responseGeminiChat2OpenAI(response *GeminiChatResponse) *OpenAITextResponse
 			Index: i,
 			Message: Message{
 				Role:    "assistant",
-				Content: candidate.Content.Parts[0].Text,
+				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
@@ -167,9 +182,7 @@ func responseGeminiChat2OpenAI(response *GeminiChatResponse) *OpenAITextResponse
 
 func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) *ChatCompletionsStreamResponse {
 	var choice ChatCompletionsStreamResponseChoice
-	if len(geminiResponse.Candidates) > 0 && len(geminiResponse.Candidates[0].Content.Parts) > 0 {
-		choice.Delta.Content = geminiResponse.Candidates[0].Content.Parts[0].Text
-	}
+	choice.Delta.Content = geminiResponse.GetResponseText()
 	choice.FinishReason = &stopFinishReason
 	var response ChatCompletionsStreamResponse
 	response.Object = "chat.completion.chunk"
@@ -180,50 +193,61 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) *ChatCo
 
 func geminiChatStreamHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithStatusCode, string) {
 	responseText := ""
-	responseId := fmt.Sprintf("chatcmpl-%s", common.GetUUID())
-	createdTime := common.GetTimestamp()
 	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() {
-		responseBody, err := io.ReadAll(resp.Body)
-		if err != nil {
-			common.SysError("error reading stream response: " + err.Error())
-			stopChan <- true
-			return
+		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
 		}
-		err = resp.Body.Close()
-		if err != nil {
-			common.SysError("error closing stream response: " + err.Error())
-			stopChan <- true
-			return
-		}
-		var geminiResponse GeminiChatResponse
-		err = json.Unmarshal(responseBody, &geminiResponse)
-		if err != nil {
-			common.SysError("error unmarshalling stream response: " + err.Error())
-			stopChan <- true
-			return
-		}
-		fullTextResponse := streamResponseGeminiChat2OpenAI(&geminiResponse)
-		fullTextResponse.Id = responseId
-		fullTextResponse.Created = createdTime
-		if len(geminiResponse.Candidates) > 0 && len(geminiResponse.Candidates[0].Content.Parts) > 0 {
-			responseText += geminiResponse.Candidates[0].Content.Parts[0].Text
-		}
-		jsonResponse, err := json.Marshal(fullTextResponse)
-		if err != nil {
-			common.SysError("error marshalling stream response: " + err.Error())
-			stopChan <- true
-			return
-		}
-		dataChan <- string(jsonResponse)
 		stopChan <- true
 	}()
 	setEventStreamHeaders(c)
 	c.Stream(func(w io.Writer) bool {
 		select {
 		case data := <-dataChan:
-			c.Render(-1, common.CustomEvent{Data: "data: " + data})
+			// 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]"})
@@ -263,7 +287,7 @@ func geminiChatHandler(c *gin.Context, resp *http.Response, promptTokens int, mo
 		}, nil
 	}
 	fullTextResponse := responseGeminiChat2OpenAI(&geminiResponse)
-	completionTokens := countTokenText(geminiResponse.Candidates[0].Content.Parts[0].Text, model)
+	completionTokens := countTokenText(geminiResponse.GetResponseText(), model)
 	usage := Usage{
 		PromptTokens:     promptTokens,
 		CompletionTokens: completionTokens,

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
@@ -79,7 +82,10 @@ func relayImageHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 
 	// Number of generated images validation
 	if isWithinRange(imageModel, imageRequest.N) == false {
-		return errorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
+		// channel not azure
+		if channelType != common.ChannelTypeAzure {
+			return errorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
+		}
 	}
 
 	// map model name
@@ -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

controller/relay-text.go

@@ -58,6 +58,9 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 	if err != nil {
 		return errorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
 	}
+	if textRequest.MaxTokens < 0 || textRequest.MaxTokens > math.MaxInt32/2 {
+		return errorWrapper(errors.New("max_tokens is invalid"), "invalid_max_tokens", http.StatusBadRequest)
+	}
 	if relayMode == RelayModeModerations && textRequest.Model == "" {
 		textRequest.Model = "text-moderation-latest"
 	}
@@ -190,10 +193,9 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			version = c.GetString("api_version")
 		}
 		action := "generateContent"
-		// actually gemini does not support stream, it's fake
-		//if textRequest.Stream {
-		//	action = "streamGenerateContent"
-		//}
+		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 ")

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
 }
 

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)

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 {

model/main.go

@@ -1,6 +1,7 @@
 package model
 
 import (
+	"fmt"
 	"gorm.io/driver/mysql"
 	"gorm.io/driver/postgres"
 	"gorm.io/driver/sqlite"
@@ -59,7 +60,8 @@ func chooseDB() (*gorm.DB, error) {
 	// Use SQLite
 	common.SysLog("SQL_DSN not set, using SQLite as database")
 	common.UsingSQLite = true
-	return gorm.Open(sqlite.Open(common.SQLitePath), &gorm.Config{
+	config := fmt.Sprintf("?_busy_timeout=%d", common.SQLiteBusyTimeout)
+	return gorm.Open(sqlite.Open(common.SQLitePath+config), &gorm.Config{
 		PrepareStmt: true, // precompile SQL
 	})
 }

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'];