feat: add batch update support (close #414)

Co-authored-by: 13714733197 <13714733197@163.com>

README.md

@@ -68,6 +68,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [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)
 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] [API2D](https://api2d.com/r/197971)
@@ -108,6 +109,8 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
 
 数据将会保存在宿主机的 `/home/ubuntu/data/one-api` 目录，请确保该目录存在且具有写入权限，或者更改为合适的目录。
 
+如果启动失败，请添加 `--privileged=true`，具体参考 #482。
+
 如果上面的镜像无法拉取，可以尝试使用 GitHub 的 Docker 镜像，将上面的 `justsong/one-api` 替换为 `ghcr.io/songquanpeng/one-api` 即可。
 
 如果你的并发量较大，**务必**设置 `SQL_DSN`，详见下面[环境变量](#环境变量)一节。
@@ -274,8 +277,9 @@ graph LR
 不加的话将会使用负载均衡的方式使用多个渠道。
 
 ### 环境变量
-1. `REDIS_CONN_STRING`：设置之后将使用 Redis 作为请求频率限制的存储，而非使用内存存储。
+1. `REDIS_CONN_STRING`：设置之后将使用 Redis 作为缓存使用。
    + 例子：`REDIS_CONN_STRING=redis://default:redispw@localhost:49153`
+   + 如果数据库访问延迟很低，没有必要启用 Redis，启用后反而会出现数据滞后的问题。
 2. `SESSION_SECRET`：设置之后将使用固定的会话密钥，这样系统重新启动后已登录用户的 cookie 将依旧有效。
    + 例子：`SESSION_SECRET=random_string`
 3. `SQL_DSN`：设置之后将使用指定数据库而非 SQLite，请使用 MySQL 或 PostgreSQL。
@@ -302,6 +306,10 @@ graph LR
    + 例子：`CHANNEL_TEST_FREQUENCY=1440`
 9. `POLLING_INTERVAL`：批量更新渠道余额以及测试可用性时的请求间隔，单位为秒，默认无间隔。
    + 例子：`POLLING_INTERVAL=5`
+10. `BATCH_UPDATE_ENABLED`：启用数据库批量更新聚合，会导致用户额度的更新存在一定的延迟可选值为 `true` 和 `false`，未设置则默认为 `false`。
+    + 例子：`BATCH_UPDATE_ENABLED=true`
+11. `BATCH_UPDATE_INTERVAL=5`：批量更新聚合的时间间隔，单位为秒，默认为 `5`。
+    + 例子：`BATCH_UPDATE_INTERVAL=5`
 
 ### 命令行参数
 1. `--port <port_number>`: 指定服务器监听的端口号，默认为 `3000`。
@@ -338,6 +346,7 @@ https://openai.justsong.cn
 5. ChatGPT Next Web 报错：`Failed to fetch`
    + 部署的时候不要设置 `BASE_URL`。
    + 检查你的接口地址和 API Key 有没有填对。
+   + 检查是否启用了 HTTPS，浏览器会拦截 HTTPS 域名下的 HTTP 请求。
 6. 报错：`当前分组负载已饱和，请稍后再试`
    + 上游通道 429 了。
 

common/constants.go

@@ -94,6 +94,9 @@ var RequestInterval = time.Duration(requestInterval) * time.Second
 
 var SyncFrequency = 10 * 60 // unit is second, will be overwritten by SYNC_FREQUENCY
 
+var BatchUpdateEnabled = false
+var BatchUpdateInterval = GetOrDefault("BATCH_UPDATE_INTERVAL", 5)
+
 const (
 	RoleGuestUser  = 0
 	RoleCommonUser = 1
@@ -173,6 +176,9 @@ const (
 	ChannelTypeZhipu          = 16
 	ChannelTypeAli            = 17
 	ChannelTypeXunfei         = 18
+	ChannelType360            = 19
+	ChannelTypeOpenRouter     = 20
+	ChannelTypeAIProxyLibrary = 21
 )
 
 var ChannelBaseURLs = []string{
@@ -195,4 +201,7 @@ var ChannelBaseURLs = []string{
 	"https://open.bigmodel.cn",       // 16
 	"https://dashscope.aliyuncs.com", // 17
 	"",                               // 18
+	"https://ai.360.cn",              // 19
+	"https://openrouter.ai/api",      // 20
+	"https://api.aiproxy.io",         // 21
 }

common/model-ratio.go

@@ -31,7 +31,7 @@ var ModelRatio = map[string]float64{
 	"text-davinci-003":          10,
 	"text-davinci-edit-001":     10,
 	"code-davinci-edit-001":     10,
-	"whisper-1":               10,
+	"whisper-1":                 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
 	"davinci":                   10,
 	"curie":                     10,
 	"babbage":                   10,
@@ -53,6 +53,11 @@ var ModelRatio = map[string]float64{
 	"qwen-v1":                   0.8572, // TBD: https://help.aliyun.com/document_detail/2399482.html?spm=a2c4g.2399482.0.0.1ad347feilAgag
 	"qwen-plus-v1":              0.5715, // Same as above
 	"SparkDesk":                 0.8572, // TBD
+	"360GPT_S2_V9":              0.8572, // ¥0.012 / 1k tokens
+	"embedding-bert-512-v1":     0.0715, // ¥0.001 / 1k tokens
+	"embedding_s1_v1":           0.0715, // ¥0.001 / 1k tokens
+	"semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
+	"360GPT_S2_V9.4":            0.8572, // ¥0.012 / 1k tokens
 }
 
 func ModelRatio2JSONString() string {

controller/channel-test.go

@@ -24,6 +24,10 @@ func testChannel(channel *model.Channel, request ChatRequest) (error, *OpenAIErr
 		fallthrough
 	case common.ChannelTypeZhipu:
 		fallthrough
+	case common.ChannelTypeAli:
+		fallthrough
+	case common.ChannelType360:
+		fallthrough
 	case common.ChannelTypeXunfei:
 		return errors.New("该渠道类型当前版本不支持测试，请手动测试"), nil
 	case common.ChannelTypeAzure:

controller/channel.go

@@ -85,7 +85,7 @@ func AddChannel(c *gin.Context) {
 	}
 	channel.CreatedTime = common.GetTimestamp()
 	keys := strings.Split(channel.Key, "\n")
-	channels := make([]model.Channel, 0)
+	channels := make([]model.Channel, 0, len(keys))
 	for _, key := range keys {
 		if key == "" {
 			continue

controller/model.go

@@ -63,6 +63,15 @@ func init() {
 			Root:       "dall-e",
 			Parent:     nil,
 		},
+		{
+			Id:         "whisper-1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "whisper-1",
+			Parent:     nil,
+		},
 		{
 			Id:         "gpt-3.5-turbo",
 			Object:     "model",
@@ -360,6 +369,51 @@ func init() {
 			Root:       "SparkDesk",
 			Parent:     nil,
 		},
+		{
+			Id:         "360GPT_S2_V9",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "360GPT_S2_V9",
+			Parent:     nil,
+		},
+		{
+			Id:         "embedding-bert-512-v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "embedding-bert-512-v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "embedding_s1_v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "embedding_s1_v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "semantic_similarity_s1_v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "semantic_similarity_s1_v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "360GPT_S2_V9.4",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "360GPT_S2_V9.4",
+			Parent:     nil,
+		},
 	}
 	openAIModelsMap = make(map[string]OpenAIModels)
 	for _, model := range openAIModels {

controller/relay-aiproxy.go

@@ -0,0 +1,220 @@
+package controller
+
+import (
+	"bufio"
+	"encoding/json"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"io"
+	"net/http"
+	"one-api/common"
+	"strconv"
+	"strings"
+)
+
+// https://docs.aiproxy.io/dev/library#使用已经定制好的知识库进行对话问答
+
+type AIProxyLibraryRequest struct {
+	Model     string `json:"model"`
+	Query     string `json:"query"`
+	LibraryId string `json:"libraryId"`
+	Stream    bool   `json:"stream"`
+}
+
+type AIProxyLibraryError struct {
+	ErrCode int    `json:"errCode"`
+	Message string `json:"message"`
+}
+
+type AIProxyLibraryDocument struct {
+	Title string `json:"title"`
+	URL   string `json:"url"`
+}
+
+type AIProxyLibraryResponse struct {
+	Success   bool                     `json:"success"`
+	Answer    string                   `json:"answer"`
+	Documents []AIProxyLibraryDocument `json:"documents"`
+	AIProxyLibraryError
+}
+
+type AIProxyLibraryStreamResponse struct {
+	Content   string                   `json:"content"`
+	Finish    bool                     `json:"finish"`
+	Model     string                   `json:"model"`
+	Documents []AIProxyLibraryDocument `json:"documents"`
+}
+
+func requestOpenAI2AIProxyLibrary(request GeneralOpenAIRequest) *AIProxyLibraryRequest {
+	query := ""
+	if len(request.Messages) != 0 {
+		query = request.Messages[len(request.Messages)-1].Content
+	}
+	return &AIProxyLibraryRequest{
+		Model:  request.Model,
+		Stream: request.Stream,
+		Query:  query,
+	}
+}
+
+func aiProxyDocuments2Markdown(documents []AIProxyLibraryDocument) string {
+	if len(documents) == 0 {
+		return ""
+	}
+	content := "\n\n参考文档：\n"
+	for i, document := range documents {
+		content += fmt.Sprintf("%d. [%s](%s)\n", i+1, document.Title, document.URL)
+	}
+	return content
+}
+
+func responseAIProxyLibrary2OpenAI(response *AIProxyLibraryResponse) *OpenAITextResponse {
+	content := response.Answer + aiProxyDocuments2Markdown(response.Documents)
+	choice := OpenAITextResponseChoice{
+		Index: 0,
+		Message: Message{
+			Role:    "assistant",
+			Content: content,
+		},
+		FinishReason: "stop",
+	}
+	fullTextResponse := OpenAITextResponse{
+		Id:      common.GetUUID(),
+		Object:  "chat.completion",
+		Created: common.GetTimestamp(),
+		Choices: []OpenAITextResponseChoice{choice},
+	}
+	return &fullTextResponse
+}
+
+func documentsAIProxyLibrary(documents []AIProxyLibraryDocument) *ChatCompletionsStreamResponse {
+	var choice ChatCompletionsStreamResponseChoice
+	choice.Delta.Content = aiProxyDocuments2Markdown(documents)
+	choice.FinishReason = &stopFinishReason
+	return &ChatCompletionsStreamResponse{
+		Id:      common.GetUUID(),
+		Object:  "chat.completion.chunk",
+		Created: common.GetTimestamp(),
+		Model:   "",
+		Choices: []ChatCompletionsStreamResponseChoice{choice},
+	}
+}
+
+func streamResponseAIProxyLibrary2OpenAI(response *AIProxyLibraryStreamResponse) *ChatCompletionsStreamResponse {
+	var choice ChatCompletionsStreamResponseChoice
+	choice.Delta.Content = response.Content
+	return &ChatCompletionsStreamResponse{
+		Id:      common.GetUUID(),
+		Object:  "chat.completion.chunk",
+		Created: common.GetTimestamp(),
+		Model:   response.Model,
+		Choices: []ChatCompletionsStreamResponseChoice{choice},
+	}
+}
+
+func aiProxyLibraryStreamHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithStatusCode, *Usage) {
+	var usage Usage
+	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
+	})
+	dataChan := make(chan string)
+	stopChan := make(chan bool)
+	go func() {
+		for scanner.Scan() {
+			data := scanner.Text()
+			if len(data) < 5 { // ignore blank line or wrong format
+				continue
+			}
+			if data[:5] != "data:" {
+				continue
+			}
+			data = data[5:]
+			dataChan <- data
+		}
+		stopChan <- true
+	}()
+	setEventStreamHeaders(c)
+	var documents []AIProxyLibraryDocument
+	c.Stream(func(w io.Writer) bool {
+		select {
+		case data := <-dataChan:
+			var AIProxyLibraryResponse AIProxyLibraryStreamResponse
+			err := json.Unmarshal([]byte(data), &AIProxyLibraryResponse)
+			if err != nil {
+				common.SysError("error unmarshalling stream response: " + err.Error())
+				return true
+			}
+			if len(AIProxyLibraryResponse.Documents) != 0 {
+				documents = AIProxyLibraryResponse.Documents
+			}
+			response := streamResponseAIProxyLibrary2OpenAI(&AIProxyLibraryResponse)
+			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:
+			response := documentsAIProxyLibrary(documents)
+			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)})
+			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), nil
+	}
+	return nil, &usage
+}
+
+func aiProxyLibraryHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithStatusCode, *Usage) {
+	var AIProxyLibraryResponse AIProxyLibraryResponse
+	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
+	}
+	err = json.Unmarshal(responseBody, &AIProxyLibraryResponse)
+	if err != nil {
+		return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	if AIProxyLibraryResponse.ErrCode != 0 {
+		return &OpenAIErrorWithStatusCode{
+			OpenAIError: OpenAIError{
+				Message: AIProxyLibraryResponse.Message,
+				Type:    strconv.Itoa(AIProxyLibraryResponse.ErrCode),
+				Code:    AIProxyLibraryResponse.ErrCode,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+	fullTextResponse := responseAIProxyLibrary2OpenAI(&AIProxyLibraryResponse)
+	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, &fullTextResponse.Usage
+}

controller/relay-audio.go

@@ -0,0 +1,147 @@
+package controller
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"one-api/common"
+	"one-api/model"
+
+	"github.com/gin-gonic/gin"
+)
+
+func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
+	audioModel := "whisper-1"
+
+	tokenId := c.GetInt("token_id")
+	channelType := c.GetInt("channel")
+	userId := c.GetInt("id")
+	group := c.GetString("group")
+
+	preConsumedTokens := common.PreConsumedQuota
+	modelRatio := common.GetModelRatio(audioModel)
+	groupRatio := common.GetGroupRatio(group)
+	ratio := modelRatio * groupRatio
+	preConsumedQuota := int(float64(preConsumedTokens) * ratio)
+	userQuota, err := model.CacheGetUserQuota(userId)
+	if err != nil {
+		return errorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError)
+	}
+	err = model.CacheDecreaseUserQuota(userId, preConsumedQuota)
+	if err != nil {
+		return errorWrapper(err, "decrease_user_quota_failed", http.StatusInternalServerError)
+	}
+	if userQuota > 100*preConsumedQuota {
+		// in this case, we do not pre-consume quota
+		// because the user has enough quota
+		preConsumedQuota = 0
+	}
+	if preConsumedQuota > 0 {
+		err := model.PreConsumeTokenQuota(tokenId, preConsumedQuota)
+		if err != nil {
+			return errorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden)
+		}
+	}
+
+	// map model name
+	modelMapping := c.GetString("model_mapping")
+	if modelMapping != "" {
+		modelMap := make(map[string]string)
+		err := json.Unmarshal([]byte(modelMapping), &modelMap)
+		if err != nil {
+			return errorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
+		}
+		if modelMap[audioModel] != "" {
+			audioModel = modelMap[audioModel]
+		}
+	}
+
+	baseURL := common.ChannelBaseURLs[channelType]
+	requestURL := c.Request.URL.String()
+
+	if c.GetString("base_url") != "" {
+		baseURL = c.GetString("base_url")
+	}
+
+	fullRequestURL := fmt.Sprintf("%s%s", baseURL, requestURL)
+	requestBody := c.Request.Body
+
+	req, err := http.NewRequest(c.Request.Method, fullRequestURL, requestBody)
+	if err != nil {
+		return errorWrapper(err, "new_request_failed", http.StatusInternalServerError)
+	}
+	req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
+	req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
+	req.Header.Set("Accept", c.Request.Header.Get("Accept"))
+
+	resp, err := httpClient.Do(req)
+	if err != nil {
+		return errorWrapper(err, "do_request_failed", http.StatusInternalServerError)
+	}
+
+	err = req.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
+	}
+	err = c.Request.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
+	}
+	var audioResponse AudioResponse
+
+	defer func() {
+		go func() {
+			quota := countTokenText(audioResponse.Text, audioModel)
+			quotaDelta := quota - preConsumedQuota
+			err := model.PostConsumeTokenQuota(tokenId, quotaDelta)
+			if err != nil {
+				common.SysError("error consuming token remain quota: " + err.Error())
+			}
+			err = model.CacheUpdateUserQuota(userId)
+			if err != nil {
+				common.SysError("error update user quota cache: " + err.Error())
+			}
+			if quota != 0 {
+				tokenName := c.GetString("token_name")
+				logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
+				model.RecordConsumeLog(userId, 0, 0, audioModel, tokenName, quota, logContent)
+				model.UpdateUserUsedQuotaAndRequestCount(userId, quota)
+				channelId := c.GetInt("channel_id")
+				model.UpdateChannelUsedQuota(channelId, quota)
+			}
+		}()
+	}()
+
+	responseBody, err := io.ReadAll(resp.Body)
+
+	if err != nil {
+		return errorWrapper(err, "read_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	err = json.Unmarshal(responseBody, &audioResponse)
+	if err != nil {
+		return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+	}
+
+	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
+
+	for k, v := range resp.Header {
+		c.Writer.Header().Set(k, v[0])
+	}
+	c.Writer.WriteHeader(resp.StatusCode)
+
+	_, err = io.Copy(c.Writer, resp.Body)
+	if err != nil {
+		return errorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	return nil
+}

controller/relay-text.go

@@ -22,6 +22,7 @@ const (
 	APITypeZhipu
 	APITypeAli
 	APITypeXunfei
+	APITypeAIProxyLibrary
 )
 
 var httpClient *http.Client
@@ -104,6 +105,8 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		apiType = APITypeAli
 	case common.ChannelTypeXunfei:
 		apiType = APITypeXunfei
+	case common.ChannelTypeAIProxyLibrary:
+		apiType = APITypeAIProxyLibrary
 	}
 	baseURL := common.ChannelBaseURLs[channelType]
 	requestURL := c.Request.URL.String()
@@ -171,6 +174,8 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		fullRequestURL = fmt.Sprintf("https://open.bigmodel.cn/api/paas/v3/model-api/%s/%s", textRequest.Model, method)
 	case APITypeAli:
 		fullRequestURL = "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation"
+	case APITypeAIProxyLibrary:
+		fullRequestURL = fmt.Sprintf("%s/api/library/ask", baseURL)
 	}
 	var promptTokens int
 	var completionTokens int
@@ -263,6 +268,14 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			return errorWrapper(err, "marshal_text_request_failed", http.StatusInternalServerError)
 		}
 		requestBody = bytes.NewBuffer(jsonStr)
+	case APITypeAIProxyLibrary:
+		aiProxyLibraryRequest := requestOpenAI2AIProxyLibrary(textRequest)
+		aiProxyLibraryRequest.LibraryId = c.GetString("library_id")
+		jsonStr, err := json.Marshal(aiProxyLibraryRequest)
+		if err != nil {
+			return errorWrapper(err, "marshal_text_request_failed", http.StatusInternalServerError)
+		}
+		requestBody = bytes.NewBuffer(jsonStr)
 	}
 
 	var req *http.Request
@@ -282,6 +295,10 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 				req.Header.Set("api-key", apiKey)
 			} else {
 				req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
+				if channelType == common.ChannelTypeOpenRouter {
+					req.Header.Set("HTTP-Referer", "https://github.com/songquanpeng/one-api")
+					req.Header.Set("X-Title", "One API")
+				}
 			}
 		case APITypeClaude:
 			req.Header.Set("x-api-key", apiKey)
@@ -298,6 +315,8 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			if textRequest.Stream {
 				req.Header.Set("X-DashScope-SSE", "enable")
 			}
+		default:
+			req.Header.Set("Authorization", "Bearer "+apiKey)
 		}
 		req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
 		req.Header.Set("Accept", c.Request.Header.Get("Accept"))
@@ -512,6 +531,26 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		} else {
 			return errorWrapper(errors.New("xunfei api does not support non-stream mode"), "invalid_api_type", http.StatusBadRequest)
 		}
+	case APITypeAIProxyLibrary:
+		if isStream {
+			err, usage := aiProxyLibraryStreamHandler(c, resp)
+			if err != nil {
+				return err
+			}
+			if usage != nil {
+				textResponse.Usage = *usage
+			}
+			return nil
+		} else {
+			err, usage := aiProxyLibraryHandler(c, resp)
+			if err != nil {
+				return err
+			}
+			if usage != nil {
+				textResponse.Usage = *usage
+			}
+			return nil
+		}
 	default:
 		return errorWrapper(errors.New("unknown api type"), "unknown_api_type", http.StatusInternalServerError)
 	}

controller/relay-utils.go

@@ -15,6 +15,24 @@ var stopFinishReason = "stop"
 
 var tokenEncoderMap = map[string]*tiktoken.Tiktoken{}
 
+func InitTokenEncoders() {
+	common.SysLog("initializing token encoders")
+	fallbackTokenEncoder, err := tiktoken.EncodingForModel("gpt-3.5-turbo")
+	if err != nil {
+		common.FatalLog(fmt.Sprintf("failed to get fallback token encoder: %s", err.Error()))
+	}
+	for model, _ := range common.ModelRatio {
+		tokenEncoder, err := tiktoken.EncodingForModel(model)
+		if err != nil {
+			common.SysError(fmt.Sprintf("using fallback encoder for model %s", model))
+			tokenEncoderMap[model] = fallbackTokenEncoder
+			continue
+		}
+		tokenEncoderMap[model] = tokenEncoder
+	}
+	common.SysLog("token encoders initialized")
+}
+
 func getTokenEncoder(model string) *tiktoken.Tiktoken {
 	if tokenEncoder, ok := tokenEncoderMap[model]; ok {
 		return tokenEncoder

controller/relay.go

@@ -24,6 +24,7 @@ const (
 	RelayModeModerations
 	RelayModeImagesGenerations
 	RelayModeEdits
+	RelayModeAudio
 )
 
 // https://platform.openai.com/docs/api-reference/chat
@@ -40,6 +41,7 @@ type GeneralOpenAIRequest struct {
 	Input       any       `json:"input,omitempty"`
 	Instruction string    `json:"instruction,omitempty"`
 	Size        string    `json:"size,omitempty"`
+	Functions   any       `json:"functions,omitempty"`
 }
 
 type ChatRequest struct {
@@ -62,6 +64,10 @@ type ImageRequest struct {
 	Size   string `json:"size"`
 }
 
+type AudioResponse struct {
+	Text string `json:"text,omitempty"`
+}
+
 type Usage struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
@@ -158,11 +164,15 @@ func Relay(c *gin.Context) {
 		relayMode = RelayModeImagesGenerations
 	} else if strings.HasPrefix(c.Request.URL.Path, "/v1/edits") {
 		relayMode = RelayModeEdits
+	} else if strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+		relayMode = RelayModeAudio
 	}
 	var err *OpenAIErrorWithStatusCode
 	switch relayMode {
 	case RelayModeImagesGenerations:
 		err = relayImageHelper(c, relayMode)
+	case RelayModeAudio:
+		err = relayAudioHelper(c, relayMode)
 	default:
 		err = relayTextHelper(c, relayMode)
 	}

main.go

@@ -77,6 +77,12 @@ func main() {
 		}
 		go controller.AutomaticallyTestChannels(frequency)
 	}
+	if os.Getenv("BATCH_UPDATE_ENABLED") == "true" {
+		common.BatchUpdateEnabled = true
+		common.SysLog("batch update enabled with interval " + strconv.Itoa(common.BatchUpdateInterval) + "s")
+		model.InitBatchUpdater()
+	}
+	controller.InitTokenEncoders()
 
 	// Initialize HTTP server
 	server := gin.Default()

middleware/distributor.go

@@ -58,7 +58,10 @@ func Distribute() func(c *gin.Context) {
 		} else {
 			// Select a channel for the user
 			var modelRequest ModelRequest
-			err := common.UnmarshalBodyReusable(c, &modelRequest)
+			var err error
+			if !strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+				err = common.UnmarshalBodyReusable(c, &modelRequest)
+			}
 			if err != nil {
 				c.JSON(http.StatusBadRequest, gin.H{
 					"error": gin.H{
@@ -84,6 +87,11 @@ func Distribute() func(c *gin.Context) {
 					modelRequest.Model = "dall-e"
 				}
 			}
+			if strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+				if modelRequest.Model == "" {
+					modelRequest.Model = "whisper-1"
+				}
+			}
 			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model)
 			if err != nil {
 				message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", userGroup, modelRequest.Model)
@@ -107,8 +115,13 @@ func Distribute() func(c *gin.Context) {
 		c.Set("model_mapping", channel.ModelMapping)
 		c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))
 		c.Set("base_url", channel.BaseURL)
-		if channel.Type == common.ChannelTypeAzure || channel.Type == common.ChannelTypeXunfei {
+		switch channel.Type {
+		case common.ChannelTypeAzure:
 			c.Set("api_version", channel.Other)
+		case common.ChannelTypeXunfei:
+			c.Set("api_version", channel.Other)
+		case common.ChannelTypeAIProxyLibrary:
+			c.Set("library_id", channel.Other)
 		}
 		c.Next()
 	}

model/channel.go

@@ -141,6 +141,14 @@ func UpdateChannelStatusById(id int, status int) {
 }
 
 func UpdateChannelUsedQuota(id int, quota int) {
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeChannelUsedQuota, id, quota)
+		return
+	}
+	updateChannelUsedQuota(id, quota)
+}
+
+func updateChannelUsedQuota(id int, quota int) {
 	err := DB.Model(&Channel{}).Where("id = ?", id).Update("used_quota", gorm.Expr("used_quota + ?", quota)).Error
 	if err != nil {
 		common.SysError("failed to update channel used quota: " + err.Error())

model/token.go

@@ -131,6 +131,14 @@ func IncreaseTokenQuota(id int, quota int) (err error) {
 	if quota < 0 {
 		return errors.New("quota 不能为负数！")
 	}
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeTokenQuota, id, quota)
+		return nil
+	}
+	return increaseTokenQuota(id, quota)
+}
+
+func increaseTokenQuota(id int, quota int) (err error) {
 	err = DB.Model(&Token{}).Where("id = ?", id).Updates(
 		map[string]interface{}{
 			"remain_quota": gorm.Expr("remain_quota + ?", quota),
@@ -144,6 +152,14 @@ func DecreaseTokenQuota(id int, quota int) (err error) {
 	if quota < 0 {
 		return errors.New("quota 不能为负数！")
 	}
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeTokenQuota, id, -quota)
+		return nil
+	}
+	return decreaseTokenQuota(id, quota)
+}
+
+func decreaseTokenQuota(id int, quota int) (err error) {
 	err = DB.Model(&Token{}).Where("id = ?", id).Updates(
 		map[string]interface{}{
 			"remain_quota": gorm.Expr("remain_quota - ?", quota),

model/user.go

@@ -275,6 +275,14 @@ func IncreaseUserQuota(id int, quota int) (err error) {
 	if quota < 0 {
 		return errors.New("quota 不能为负数！")
 	}
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeUserQuota, id, quota)
+		return nil
+	}
+	return increaseUserQuota(id, quota)
+}
+
+func increaseUserQuota(id int, quota int) (err error) {
 	err = DB.Model(&User{}).Where("id = ?", id).Update("quota", gorm.Expr("quota + ?", quota)).Error
 	return err
 }
@@ -283,6 +291,14 @@ func DecreaseUserQuota(id int, quota int) (err error) {
 	if quota < 0 {
 		return errors.New("quota 不能为负数！")
 	}
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeUserQuota, id, -quota)
+		return nil
+	}
+	return decreaseUserQuota(id, quota)
+}
+
+func decreaseUserQuota(id int, quota int) (err error) {
 	err = DB.Model(&User{}).Where("id = ?", id).Update("quota", gorm.Expr("quota - ?", quota)).Error
 	return err
 }
@@ -293,10 +309,18 @@ func GetRootUserEmail() (email string) {
 }
 
 func UpdateUserUsedQuotaAndRequestCount(id int, quota int) {
+	if common.BatchUpdateEnabled {
+		addNewRecord(BatchUpdateTypeUsedQuotaAndRequestCount, id, quota)
+		return
+	}
+	updateUserUsedQuotaAndRequestCount(id, quota, 1)
+}
+
+func updateUserUsedQuotaAndRequestCount(id int, quota int, count int) {
 	err := DB.Model(&User{}).Where("id = ?", id).Updates(
 		map[string]interface{}{
 			"used_quota":    gorm.Expr("used_quota + ?", quota),
-			"request_count": gorm.Expr("request_count + ?", 1),
+			"request_count": gorm.Expr("request_count + ?", count),
 		},
 	).Error
 	if err != nil {

model/utils.go

@@ -0,0 +1,75 @@
+package model
+
+import (
+	"one-api/common"
+	"sync"
+	"time"
+)
+
+const BatchUpdateTypeCount = 4 // if you add a new type, you need to add a new map and a new lock
+
+const (
+	BatchUpdateTypeUserQuota = iota
+	BatchUpdateTypeTokenQuota
+	BatchUpdateTypeUsedQuotaAndRequestCount
+	BatchUpdateTypeChannelUsedQuota
+)
+
+var batchUpdateStores []map[int]int
+var batchUpdateLocks []sync.Mutex
+
+func init() {
+	for i := 0; i < BatchUpdateTypeCount; i++ {
+		batchUpdateStores = append(batchUpdateStores, make(map[int]int))
+		batchUpdateLocks = append(batchUpdateLocks, sync.Mutex{})
+	}
+}
+
+func InitBatchUpdater() {
+	go func() {
+		for {
+			time.Sleep(time.Duration(common.BatchUpdateInterval) * time.Second)
+			batchUpdate()
+		}
+	}()
+}
+
+func addNewRecord(type_ int, id int, value int) {
+	batchUpdateLocks[type_].Lock()
+	defer batchUpdateLocks[type_].Unlock()
+	if _, ok := batchUpdateStores[type_][id]; !ok {
+		batchUpdateStores[type_][id] = value
+	} else {
+		batchUpdateStores[type_][id] += value
+	}
+}
+
+func batchUpdate() {
+	common.SysLog("batch update started")
+	for i := 0; i < BatchUpdateTypeCount; i++ {
+		batchUpdateLocks[i].Lock()
+		store := batchUpdateStores[i]
+		batchUpdateStores[i] = make(map[int]int)
+		batchUpdateLocks[i].Unlock()
+
+		for key, value := range store {
+			switch i {
+			case BatchUpdateTypeUserQuota:
+				err := increaseUserQuota(key, value)
+				if err != nil {
+					common.SysError("failed to batch update user quota: " + err.Error())
+				}
+			case BatchUpdateTypeTokenQuota:
+				err := increaseTokenQuota(key, value)
+				if err != nil {
+					common.SysError("failed to batch update token quota: " + err.Error())
+				}
+			case BatchUpdateTypeUsedQuotaAndRequestCount:
+				updateUserUsedQuotaAndRequestCount(key, value, 1) // TODO: count is incorrect
+			case BatchUpdateTypeChannelUsedQuota:
+				updateChannelUsedQuota(key, value)
+			}
+		}
+	}
+	common.SysLog("batch update finished")
+}

router/relay-router.go

@@ -26,8 +26,8 @@ func SetRelayRouter(router *gin.Engine) {
 		relayV1Router.POST("/images/variations", controller.RelayNotImplemented)
 		relayV1Router.POST("/embeddings", controller.Relay)
 		relayV1Router.POST("/engines/:model/embeddings", controller.Relay)
-		relayV1Router.POST("/audio/transcriptions", controller.RelayNotImplemented)
+		relayV1Router.POST("/audio/transcriptions", controller.Relay)
-		relayV1Router.POST("/audio/translations", controller.RelayNotImplemented)
+		relayV1Router.POST("/audio/translations", controller.Relay)
 		relayV1Router.GET("/files", controller.RelayNotImplemented)
 		relayV1Router.POST("/files", controller.RelayNotImplemented)
 		relayV1Router.DELETE("/files/:id", controller.RelayNotImplemented)

web/src/components/LogsTable.js

@@ -324,7 +324,7 @@ const LogsTable = () => {
               .map((log, idx) => {
                 if (log.deleted) return <></>;
                 return (
-                  <Table.Row key={log.created_at}>
+                  <Table.Row key={log.id}>
                     <Table.Cell>{renderTimestamp(log.created_at)}</Table.Cell>
                     {
                       isAdminUser && (

web/src/constants/channel.constants.js

@@ -7,7 +7,10 @@ export const CHANNEL_OPTIONS = [
   { 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' },
   { key: 16, text: '智谱 ChatGLM', value: 16, color: 'violet' },
+  { key: 19, text: '360 智脑', value: 19, color: 'blue' },
   { key: 8, text: '自定义渠道', value: 8, color: 'pink' },
+  { key: 21, text: '知识库：AI Proxy', value: 21, color: 'purple' },
+  { key: 20, text: '代理：OpenRouter', value: 20, color: 'black' },
   { key: 2, text: '代理：API2D', value: 2, color: 'blue' },
   { key: 5, text: '代理：OpenAI-SB', value: 5, color: 'brown' },
   { key: 7, text: '代理：OhMyGPT', value: 7, color: 'purple' },

web/src/pages/Channel/EditChannel.js

@@ -1,6 +1,6 @@
 import React, { useEffect, useState } from 'react';
 import { Button, Form, Header, Input, Message, Segment } from 'semantic-ui-react';
-import { useParams, useNavigate } from 'react-router-dom';
+import { useNavigate, useParams } from 'react-router-dom';
 import { API, showError, showInfo, showSuccess, verifyJSON } from '../../helpers';
 import { CHANNEL_OPTIONS } from '../../constants';
 
@@ -61,6 +61,9 @@ const EditChannel = () => {
         case 18:
           localModels = ['SparkDesk'];
           break;
+        case 19:
+          localModels = ['360GPT_S2_V9', 'embedding-bert-512-v1', 'embedding_s1_v1', 'semantic_similarity_s1_v1', '360GPT_S2_V9.4'];
+          break;
       }
       setInputs((inputs) => ({ ...inputs, models: localModels }));
     }
@@ -292,6 +295,20 @@ const EditChannel = () => {
               </Form.Field>
             )
           }
+          {
+            inputs.type === 21 && (
+              <Form.Field>
+                <Form.Input
+                  label='知识库 ID'
+                  name='other'
+                  placeholder={'请输入知识库 ID，例如：123456'}
+                  onChange={handleInputChange}
+                  value={inputs.other}
+                  autoComplete='new-password'
+                />
+              </Form.Field>
+            )
+          }
           <Form.Field>
             <Form.Dropdown
               label='模型'