refactor: split RelayTextHelper function

* add new 0125 chat models and embedding-3 models

* refine the step of manually deploying

* add gpt-4-turbo-preview

README.en.md

@@ -134,12 +134,12 @@ The initial account username is `root` and password is `123456`.
    git clone https://github.com/songquanpeng/one-api.git
    
    # Build the frontend
-   cd one-api/web
+   cd one-api/web/default
    npm install
    npm run build
    
    # Build the backend
-   cd ..
+   cd ../..
    go mod download
    go build -ldflags "-s -w" -o one-api
    ```

README.ja.md

@@ -135,12 +135,12 @@ sudo service nginx restart
    git clone https://github.com/songquanpeng/one-api.git
 
    # フロントエンドのビルド
-   cd one-api/web
+   cd one-api/web/default
    npm install
    npm run build
 
    # バックエンドのビルド
-   cd ..
+   cd ../..
    go mod download
    go build -ldflags "-s -w" -o one-api
    ```

README.md

@@ -174,12 +174,12 @@ docker-compose ps
    git clone https://github.com/songquanpeng/one-api.git
    
    # 构建前端
-   cd one-api/web
+   cd one-api/web/default
    npm install
    npm run build
    
    # 构建后端
-   cd ..
+   cd ../..
    go mod download
    go build -ldflags "-s -w" -o one-api
    ````

common/model-ratio.go

@@ -45,6 +45,8 @@ var ModelRatio = map[string]float64{
 	"gpt-4-32k-0314":            30,
 	"gpt-4-32k-0613":            30,
 	"gpt-4-1106-preview":        5,    // $0.01 / 1K tokens
+	"gpt-4-0125-preview":        5,    // $0.01 / 1K tokens
+	"gpt-4-turbo-preview":       5,    // $0.01 / 1K tokens
 	"gpt-4-vision-preview":      5,    // $0.01 / 1K tokens
 	"gpt-3.5-turbo":             0.75, // $0.0015 / 1K tokens
 	"gpt-3.5-turbo-0301":        0.75,
@@ -53,6 +55,7 @@ var ModelRatio = map[string]float64{
 	"gpt-3.5-turbo-16k-0613":    1.5,
 	"gpt-3.5-turbo-instruct":    0.75, // $0.0015 / 1K tokens
 	"gpt-3.5-turbo-1106":        0.5,  // $0.001 / 1K tokens
+	"gpt-3.5-turbo-0125":        0.25, // $0.0005 / 1K tokens
 	"davinci-002":               1,    // $0.002 / 1K tokens
 	"babbage-002":               0.2,  // $0.0004 / 1K tokens
 	"text-ada-001":              0.2,
@@ -72,6 +75,8 @@ var ModelRatio = map[string]float64{
 	"babbage":                   10,
 	"ada":                       10,
 	"text-embedding-ada-002":    0.05,
+	"text-embedding-3-small":    0.01,
+	"text-embedding-3-large":    0.065,
 	"text-search-ada-doc-001":   10,
 	"text-moderation-stable":    0.1,
 	"text-moderation-latest":    0.1,
@@ -130,8 +135,31 @@ func GetModelRatio(name string) float64 {
 	return ratio
 }
 
+var CompletionRatio = map[string]float64{}
+
+func CompletionRatio2JSONString() string {
+	jsonBytes, err := json.Marshal(CompletionRatio)
+	if err != nil {
+		logger.SysError("error marshalling completion ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+func UpdateCompletionRatioByJSONString(jsonStr string) error {
+	CompletionRatio = make(map[string]float64)
+	return json.Unmarshal([]byte(jsonStr), &CompletionRatio)
+}
+
 func GetCompletionRatio(name string) float64 {
+	if ratio, ok := CompletionRatio[name]; ok {
+		return ratio
+	}
 	if strings.HasPrefix(name, "gpt-3.5") {
+		if strings.HasSuffix(name, "0125") {
+			// https://openai.com/blog/new-embedding-models-and-api-updates
+			// Updated GPT-3.5 Turbo model and lower pricing
+			return 3
+		}
 		if strings.HasSuffix(name, "1106") {
 			return 2
 		}

controller/model.go

@@ -171,6 +171,15 @@ func init() {
 			Root:       "gpt-3.5-turbo-1106",
 			Parent:     nil,
 		},
+		{
+			Id:         "gpt-3.5-turbo-0125",
+			Object:     "model",
+			Created:    1706232090,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "gpt-3.5-turbo-0125",
+			Parent:     nil,
+		},
 		{
 			Id:         "gpt-3.5-turbo-instruct",
 			Object:     "model",
@@ -243,6 +252,24 @@ func init() {
 			Root:       "gpt-4-1106-preview",
 			Parent:     nil,
 		},
+		{
+			Id:         "gpt-4-0125-preview",
+			Object:     "model",
+			Created:    1706232090,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "gpt-4-0125-preview",
+			Parent:     nil,
+		},
+		{
+			Id:         "gpt-4-turbo-preview",
+			Object:     "model",
+			Created:    1706232090,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "gpt-4-turbo-preview",
+			Parent:     nil,
+		},
 		{
 			Id:         "gpt-4-vision-preview",
 			Object:     "model",
@@ -261,6 +288,24 @@ func init() {
 			Root:       "text-embedding-ada-002",
 			Parent:     nil,
 		},
+		{
+			Id:         "text-embedding-3-small",
+			Object:     "model",
+			Created:    1706232090,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "text-embedding-3-small",
+			Parent:     nil,
+		},
+		{
+			Id:         "text-embedding-3-large",
+			Object:     "model",
+			Created:    1706232090,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "text-embedding-3-large",
+			Parent:     nil,
+		},
 		{
 			Id:         "text-davinci-003",
 			Object:     "model",

controller/relay.go

@@ -29,7 +29,7 @@ func Relay(c *gin.Context) {
 	case constant.RelayModeAudioTranscription:
 		err = controller.RelayAudioHelper(c, relayMode)
 	default:
-		err = controller.RelayTextHelper(c, relayMode)
+		err = controller.RelayTextHelper(c)
 	}
 	if err != nil {
 		requestId := c.GetString(logger.RequestIdKey)

model/option.go

@@ -66,6 +66,7 @@ func InitOptionMap() {
 	config.OptionMap["PreConsumedQuota"] = strconv.Itoa(config.PreConsumedQuota)
 	config.OptionMap["ModelRatio"] = common.ModelRatio2JSONString()
 	config.OptionMap["GroupRatio"] = common.GroupRatio2JSONString()
+	config.OptionMap["CompletionRatio"] = common.CompletionRatio2JSONString()
 	config.OptionMap["TopUpLink"] = config.TopUpLink
 	config.OptionMap["ChatLink"] = config.ChatLink
 	config.OptionMap["QuotaPerUnit"] = strconv.FormatFloat(config.QuotaPerUnit, 'f', -1, 64)
@@ -198,6 +199,8 @@ func updateOptionMap(key string, value string) (err error) {
 		err = common.UpdateModelRatioByJSONString(value)
 	case "GroupRatio":
 		err = common.UpdateGroupRatioByJSONString(value)
+	case "CompletionRatio":
+		err = common.UpdateCompletionRatioByJSONString(value)
 	case "TopUpLink":
 		config.TopUpLink = value
 	case "ChatLink":

relay/controller/helper.go

@@ -0,0 +1,146 @@
+package controller
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"io"
+	"math"
+	"net/http"
+	"one-api/common"
+	"one-api/common/config"
+	"one-api/common/logger"
+	"one-api/model"
+	"one-api/relay/channel/openai"
+	"one-api/relay/constant"
+	"one-api/relay/util"
+)
+
+func getAndValidateTextRequest(c *gin.Context, relayMode int) (*openai.GeneralOpenAIRequest, error) {
+	textRequest := &openai.GeneralOpenAIRequest{}
+	err := common.UnmarshalBodyReusable(c, textRequest)
+	if err != nil {
+		return nil, err
+	}
+	if relayMode == constant.RelayModeModerations && textRequest.Model == "" {
+		textRequest.Model = "text-moderation-latest"
+	}
+	if relayMode == constant.RelayModeEmbeddings && textRequest.Model == "" {
+		textRequest.Model = c.Param("model")
+	}
+	err = util.ValidateTextRequest(textRequest, relayMode)
+	if err != nil {
+		return nil, err
+	}
+	return textRequest, nil
+}
+
+func getPromptTokens(textRequest *openai.GeneralOpenAIRequest, relayMode int) int {
+	switch relayMode {
+	case constant.RelayModeChatCompletions:
+		return openai.CountTokenMessages(textRequest.Messages, textRequest.Model)
+	case constant.RelayModeCompletions:
+		return openai.CountTokenInput(textRequest.Prompt, textRequest.Model)
+	case constant.RelayModeModerations:
+		return openai.CountTokenInput(textRequest.Input, textRequest.Model)
+	}
+	return 0
+}
+
+func getPreConsumedQuota(textRequest *openai.GeneralOpenAIRequest, promptTokens int, ratio float64) int {
+	preConsumedTokens := config.PreConsumedQuota
+	if textRequest.MaxTokens != 0 {
+		preConsumedTokens = promptTokens + textRequest.MaxTokens
+	}
+	return int(float64(preConsumedTokens) * ratio)
+}
+
+func preConsumeQuota(ctx context.Context, textRequest *openai.GeneralOpenAIRequest, promptTokens int, ratio float64, meta *util.RelayMeta) (int, *openai.ErrorWithStatusCode) {
+	preConsumedQuota := getPreConsumedQuota(textRequest, promptTokens, ratio)
+
+	userQuota, err := model.CacheGetUserQuota(meta.UserId)
+	if err != nil {
+		return preConsumedQuota, openai.ErrorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError)
+	}
+	if userQuota-preConsumedQuota < 0 {
+		return preConsumedQuota, openai.ErrorWrapper(errors.New("user quota is not enough"), "insufficient_user_quota", http.StatusForbidden)
+	}
+	err = model.CacheDecreaseUserQuota(meta.UserId, preConsumedQuota)
+	if err != nil {
+		return preConsumedQuota, openai.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
+		logger.Info(ctx, fmt.Sprintf("user %d has enough quota %d, trusted and no need to pre-consume", meta.UserId, userQuota))
+	}
+	if preConsumedQuota > 0 {
+		err := model.PreConsumeTokenQuota(meta.TokenId, preConsumedQuota)
+		if err != nil {
+			return preConsumedQuota, openai.ErrorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden)
+		}
+	}
+	return preConsumedQuota, nil
+}
+
+func postConsumeQuota(ctx context.Context, usage *openai.Usage, meta *util.RelayMeta, textRequest *openai.GeneralOpenAIRequest, ratio float64, preConsumedQuota int, modelRatio float64, groupRatio float64) {
+	if usage == nil {
+		logger.Error(ctx, "usage is nil, which is unexpected")
+		return
+	}
+	quota := 0
+	completionRatio := common.GetCompletionRatio(textRequest.Model)
+	promptTokens := usage.PromptTokens
+	completionTokens := usage.CompletionTokens
+	quota = int(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
+	if ratio != 0 && quota <= 0 {
+		quota = 1
+	}
+	totalTokens := promptTokens + completionTokens
+	if totalTokens == 0 {
+		// in this case, must be some error happened
+		// we cannot just return, because we may have to return the pre-consumed quota
+		quota = 0
+	}
+	quotaDelta := quota - preConsumedQuota
+	err := model.PostConsumeTokenQuota(meta.TokenId, quotaDelta)
+	if err != nil {
+		logger.Error(ctx, "error consuming token remain quota: "+err.Error())
+	}
+	err = model.CacheUpdateUserQuota(meta.UserId)
+	if err != nil {
+		logger.Error(ctx, "error update user quota cache: "+err.Error())
+	}
+	if quota != 0 {
+		logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
+		model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
+		model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
+		model.UpdateChannelUsedQuota(meta.ChannelId, quota)
+	}
+}
+
+func doRequest(ctx context.Context, c *gin.Context, meta *util.RelayMeta, isStream bool, fullRequestURL string, requestBody io.Reader) (*http.Response, error) {
+	req, err := http.NewRequest(c.Request.Method, fullRequestURL, requestBody)
+	if err != nil {
+		return nil, err
+	}
+	SetupRequestHeaders(c, req, meta, isStream)
+	resp, err := util.HTTPClient.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	if resp == nil {
+		return nil, errors.New("resp is nil")
+	}
+	err = req.Body.Close()
+	if err != nil {
+		logger.Warnf(ctx, "close req.Body failed: %+v", err)
+	}
+	err = c.Request.Body.Close()
+	if err != nil {
+		logger.Warnf(ctx, "close c.Request.Body failed: %+v", err)
+	}
+	return resp, nil
+}

relay/controller/temp.go

@@ -24,9 +24,9 @@ import (
 	"strings"
 )
 
-func GetRequestURL(requestURL string, apiType int, relayMode int, meta *util.RelayMeta, textRequest *openai.GeneralOpenAIRequest) (string, error) {
+func GetRequestURL(requestURL string, meta *util.RelayMeta, textRequest *openai.GeneralOpenAIRequest) (string, error) {
 	fullRequestURL := util.GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType)
-	switch apiType {
+	switch meta.APIType {
 	case constant.APITypeOpenAI:
 		if meta.ChannelType == common.ChannelTypeAzure {
 			// https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
@@ -81,7 +81,7 @@ func GetRequestURL(requestURL string, apiType int, relayMode int, meta *util.Rel
 		fullRequestURL = fmt.Sprintf("https://open.bigmodel.cn/api/paas/v3/model-api/%s/%s", textRequest.Model, method)
 	case constant.APITypeAli:
 		fullRequestURL = "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation"
-		if relayMode == constant.RelayModeEmbeddings {
+		if meta.Mode == constant.RelayModeEmbeddings {
 			fullRequestURL = "https://dashscope.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding"
 		}
 	case constant.APITypeTencent:
@@ -191,8 +191,8 @@ func GetRequestBody(c *gin.Context, textRequest openai.GeneralOpenAIRequest, isM
 	return requestBody, nil
 }
 
-func SetupRequestHeaders(c *gin.Context, req *http.Request, apiType int, meta *util.RelayMeta, isStream bool) {
-	SetupAuthHeaders(c, req, apiType, meta, isStream)
+func SetupRequestHeaders(c *gin.Context, req *http.Request, meta *util.RelayMeta, isStream bool) {
+	SetupAuthHeaders(c, req, meta, isStream)
 	req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
 	req.Header.Set("Accept", c.Request.Header.Get("Accept"))
 	if isStream && c.Request.Header.Get("Accept") == "" {
@@ -200,9 +200,9 @@ func SetupRequestHeaders(c *gin.Context, req *http.Request, apiType int, meta *u
 	}
 }
 
-func SetupAuthHeaders(c *gin.Context, req *http.Request, apiType int, meta *util.RelayMeta, isStream bool) {
+func SetupAuthHeaders(c *gin.Context, req *http.Request, meta *util.RelayMeta, isStream bool) {
 	apiKey := meta.APIKey
-	switch apiType {
+	switch meta.APIType {
 	case constant.APITypeOpenAI:
 		if meta.ChannelType == common.ChannelTypeAzure {
 			req.Header.Set("api-key", apiKey)

relay/controller/text.go

@@ -1,115 +1,61 @@
 package controller
 
 import (
-	"context"
-	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"math"
 	"net/http"
 	"one-api/common"
-	"one-api/common/config"
 	"one-api/common/logger"
-	"one-api/model"
 	"one-api/relay/channel/openai"
 	"one-api/relay/constant"
 	"one-api/relay/util"
 	"strings"
 )
 
-func RelayTextHelper(c *gin.Context, relayMode int) *openai.ErrorWithStatusCode {
+func RelayTextHelper(c *gin.Context) *openai.ErrorWithStatusCode {
 	ctx := c.Request.Context()
 	meta := util.GetRelayMeta(c)
-	var textRequest openai.GeneralOpenAIRequest
-	err := common.UnmarshalBodyReusable(c, &textRequest)
-	if err != nil {
-		return openai.ErrorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
-	}
-	if relayMode == constant.RelayModeModerations && textRequest.Model == "" {
-		textRequest.Model = "text-moderation-latest"
-	}
-	if relayMode == constant.RelayModeEmbeddings && textRequest.Model == "" {
-		textRequest.Model = c.Param("model")
-	}
-	err = util.ValidateTextRequest(&textRequest, relayMode)
+	// get & validate textRequest
+	textRequest, err := getAndValidateTextRequest(c, meta.Mode)
 	if err != nil {
+		logger.Errorf(ctx, "getAndValidateTextRequest failed: %s", err.Error())
 		return openai.ErrorWrapper(err, "invalid_text_request", http.StatusBadRequest)
 	}
+	// map model name
 	var isModelMapped bool
 	textRequest.Model, isModelMapped = util.GetMappedModelName(textRequest.Model, meta.ModelMapping)
-	apiType := constant.ChannelType2APIType(meta.ChannelType)
-	fullRequestURL, err := GetRequestURL(c.Request.URL.String(), apiType, relayMode, meta, &textRequest)
-	if err != nil {
-		logger.Error(ctx, fmt.Sprintf("util.GetRequestURL failed: %s", err.Error()))
-		return openai.ErrorWrapper(fmt.Errorf("util.GetRequestURL failed"), "get_request_url_failed", http.StatusInternalServerError)
-	}
-	var promptTokens int
-	var completionTokens int
-	switch relayMode {
-	case constant.RelayModeChatCompletions:
-		promptTokens = openai.CountTokenMessages(textRequest.Messages, textRequest.Model)
-	case constant.RelayModeCompletions:
-		promptTokens = openai.CountTokenInput(textRequest.Prompt, textRequest.Model)
-	case constant.RelayModeModerations:
-		promptTokens = openai.CountTokenInput(textRequest.Input, textRequest.Model)
-	}
-	preConsumedTokens := config.PreConsumedQuota
-	if textRequest.MaxTokens != 0 {
-		preConsumedTokens = promptTokens + textRequest.MaxTokens
-	}
+	// get model ratio & group ratio
 	modelRatio := common.GetModelRatio(textRequest.Model)
 	groupRatio := common.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
-	preConsumedQuota := int(float64(preConsumedTokens) * ratio)
-	userQuota, err := model.CacheGetUserQuota(meta.UserId)
-	if err != nil {
-		return openai.ErrorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError)
+	// pre-consume quota
+	promptTokens := getPromptTokens(textRequest, meta.Mode)
+	preConsumedQuota, bizErr := preConsumeQuota(ctx, textRequest, promptTokens, ratio, meta)
+	if bizErr != nil {
+		logger.Warnf(ctx, "preConsumeQuota failed: %+v", *bizErr)
+		return bizErr
 	}
-	if userQuota-preConsumedQuota < 0 {
-		return openai.ErrorWrapper(errors.New("user quota is not enough"), "insufficient_user_quota", http.StatusForbidden)
-	}
-	err = model.CacheDecreaseUserQuota(meta.UserId, preConsumedQuota)
-	if err != nil {
-		return openai.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
-		logger.Info(c.Request.Context(), fmt.Sprintf("user %d has enough quota %d, trusted and no need to pre-consume", meta.UserId, userQuota))
-	}
-	if preConsumedQuota > 0 {
-		err := model.PreConsumeTokenQuota(meta.TokenId, preConsumedQuota)
-		if err != nil {
-			return openai.ErrorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden)
-		}
-	}
-	requestBody, err := GetRequestBody(c, textRequest, isModelMapped, apiType, relayMode)
+
+	// get request body
+	requestBody, err := GetRequestBody(c, *textRequest, isModelMapped, meta.APIType, meta.Mode)
 	if err != nil {
 		return openai.ErrorWrapper(err, "get_request_body_failed", http.StatusInternalServerError)
 	}
-	var req *http.Request
+	// do request
 	var resp *http.Response
 	isStream := textRequest.Stream
+	if meta.APIType != constant.APITypeXunfei { // cause xunfei use websocket
+		fullRequestURL, err := GetRequestURL(c.Request.URL.String(), meta, textRequest)
+		if err != nil {
+			logger.Error(ctx, fmt.Sprintf("util.GetRequestURL failed: %s", err.Error()))
+			return openai.ErrorWrapper(fmt.Errorf("util.GetRequestURL failed"), "get_request_url_failed", http.StatusInternalServerError)
+		}
 
-	if apiType != constant.APITypeXunfei { // cause xunfei use websocket
-		req, err = http.NewRequest(c.Request.Method, fullRequestURL, requestBody)
-		if err != nil {
-			return openai.ErrorWrapper(err, "new_request_failed", http.StatusInternalServerError)
-		}
-		SetupRequestHeaders(c, req, apiType, meta, isStream)
-		resp, err = util.HTTPClient.Do(req)
+		resp, err = doRequest(ctx, c, meta, isStream, fullRequestURL, requestBody)
 		if err != nil {
+			logger.Errorf(ctx, "doRequest failed: %s", err.Error())
 			return openai.ErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
 		}
-		err = req.Body.Close()
-		if err != nil {
-			return openai.ErrorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
-		}
-		err = c.Request.Body.Close()
-		if err != nil {
-			return openai.ErrorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
-		}
 		isStream = isStream || strings.HasPrefix(resp.Header.Get("Content-Type"), "text/event-stream")
 
 		if resp.StatusCode != http.StatusOK {
@@ -117,57 +63,14 @@ func RelayTextHelper(c *gin.Context, relayMode int) *openai.ErrorWithStatusCode
 			return util.RelayErrorHandler(resp)
 		}
 	}
-
-	var respErr *openai.ErrorWithStatusCode
-	var usage *openai.Usage
-
-	defer func(ctx context.Context) {
-		// Why we use defer here? Because if error happened, we will have to return the pre-consumed quota.
-		if respErr != nil {
-			logger.Errorf(ctx, "respErr is not nil: %+v", respErr)
-			util.ReturnPreConsumedQuota(ctx, preConsumedQuota, meta.TokenId)
-			return
-		}
-		if usage == nil {
-			logger.Error(ctx, "usage is nil, which is unexpected")
-			return
-		}
-
-		go func() {
-			quota := 0
-			completionRatio := common.GetCompletionRatio(textRequest.Model)
-			promptTokens = usage.PromptTokens
-			completionTokens = usage.CompletionTokens
-			quota = int(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
-			if ratio != 0 && quota <= 0 {
-				quota = 1
-			}
-			totalTokens := promptTokens + completionTokens
-			if totalTokens == 0 {
-				// in this case, must be some error happened
-				// we cannot just return, because we may have to return the pre-consumed quota
-				quota = 0
-			}
-			quotaDelta := quota - preConsumedQuota
-			err := model.PostConsumeTokenQuota(meta.TokenId, quotaDelta)
-			if err != nil {
-				logger.Error(ctx, "error consuming token remain quota: "+err.Error())
-			}
-			err = model.CacheUpdateUserQuota(meta.UserId)
-			if err != nil {
-				logger.Error(ctx, "error update user quota cache: "+err.Error())
-			}
-			if quota != 0 {
-				logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
-				model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
-				model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
-				model.UpdateChannelUsedQuota(meta.ChannelId, quota)
-			}
-		}()
-	}(ctx)
-	usage, respErr = DoResponse(c, &textRequest, resp, relayMode, apiType, isStream, promptTokens)
+	// do response
+	usage, respErr := DoResponse(c, textRequest, resp, meta.Mode, meta.APIType, isStream, promptTokens)
 	if respErr != nil {
+		logger.Errorf(ctx, "respErr is not nil: %+v", respErr)
+		util.ReturnPreConsumedQuota(ctx, preConsumedQuota, meta.TokenId)
 		return respErr
 	}
+	// post-consume quota
+	go postConsumeQuota(ctx, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio)
 	return nil
 }

relay/util/relay_meta.go

@@ -3,10 +3,12 @@ package util
 import (
 	"github.com/gin-gonic/gin"
 	"one-api/common"
+	"one-api/relay/constant"
 	"strings"
 )
 
 type RelayMeta struct {
+	Mode         int
 	ChannelType  int
 	ChannelId    int
 	TokenId      int
@@ -17,11 +19,13 @@ type RelayMeta struct {
 	BaseURL      string
 	APIVersion   string
 	APIKey       string
+	APIType      int
 	Config       map[string]string
 }
 
 func GetRelayMeta(c *gin.Context) *RelayMeta {
 	meta := RelayMeta{
+		Mode:         constant.Path2RelayMode(c.Request.URL.Path),
 		ChannelType:  c.GetInt("channel"),
 		ChannelId:    c.GetInt("channel_id"),
 		TokenId:      c.GetInt("token_id"),
@@ -40,5 +44,6 @@ func GetRelayMeta(c *gin.Context) *RelayMeta {
 	if meta.BaseURL == "" {
 		meta.BaseURL = common.ChannelBaseURLs[meta.ChannelType]
 	}
+	meta.APIType = constant.ChannelType2APIType(meta.ChannelType)
 	return &meta
 }

web/berry/src/views/Setting/component/OperationSetting.js

@@ -27,6 +27,7 @@ const OperationSetting = () => {
     QuotaRemindThreshold: 0,
     PreConsumedQuota: 0,
     ModelRatio: "",
+    CompletionRatio: "",
     GroupRatio: "",
     TopUpLink: "",
     ChatLink: "",
@@ -52,9 +53,12 @@ const OperationSetting = () => {
     if (success) {
       let newInputs = {};
       data.forEach((item) => {
-        if (item.key === "ModelRatio" || item.key === "GroupRatio") {
+        if (item.key === "ModelRatio" || item.key === "GroupRatio" || item.key === "CompletionRatio") {
           item.value = JSON.stringify(JSON.parse(item.value), null, 2);
         }
+        if (item.value === '{}') {
+          item.value = '';
+        }
         newInputs[item.key] = item.value;
       });
       setInputs(newInputs);
@@ -133,6 +137,13 @@ const OperationSetting = () => {
           }
           await updateOption("GroupRatio", inputs.GroupRatio);
         }
+        if (originInputs['CompletionRatio'] !== inputs.CompletionRatio) {
+          if (!verifyJSON(inputs.CompletionRatio)) {
+            showError('补全倍率不是合法的 JSON 字符串');
+            return;
+          }
+          await updateOption('CompletionRatio', inputs.CompletionRatio);
+        }
         break;
       case "quota":
         if (originInputs["QuotaForNewUser"] !== inputs.QuotaForNewUser) {
@@ -500,7 +511,20 @@ const OperationSetting = () => {
               placeholder="为一个 JSON 文本，键为模型名称，值为倍率"
             />
           </FormControl>
-
+          <FormControl fullWidth>
+            <TextField
+              multiline
+              maxRows={15}
+              id="channel-CompletionRatio-label"
+              label="补全倍率"
+              value={inputs.CompletionRatio}
+              name="CompletionRatio"
+              onChange={handleInputChange}
+              aria-describedby="helper-text-channel-CompletionRatio-label"
+              minRows={5}
+              placeholder="为一个 JSON 文本，键为模型名称，值为倍率，此处的倍率设置是模型补全倍率相较于提示倍率的比例，使用该设置可强制覆盖 One API 的内部比例"
+            />
+          </FormControl>
           <FormControl fullWidth>
             <TextField
               multiline

web/berry/src/views/Token/component/TableRow.js

@@ -192,7 +192,7 @@ export default function TokensTableRow({ item, manageToken, handleOpenModal, set
               id={`switch-${item.id}`}
               checked={statusSwitch === 1}
               onChange={handleStatus}
-              // disabled={statusSwitch !== 1 && statusSwitch !== 2}
+            // disabled={statusSwitch !== 1 && statusSwitch !== 2}
             />
           </Tooltip>
         </TableCell>
@@ -222,7 +222,7 @@ export default function TokensTableRow({ item, manageToken, handleOpenModal, set
               </Button>
             </ButtonGroup>
             <ButtonGroup size="small" aria-label="split button">
-              <Button color="primary">聊天</Button>
+              <Button color="primary" onClick={(e) => handleCopy(COPY_OPTIONS[0], 'link')}>聊天</Button>
               <Button size="small" onClick={(e) => handleOpenMenu(e, 'link')}>
                 <IconCaretDownFilled size={'16px'} />
               </Button>

web/default/src/components/OperationSetting.js

@@ -11,6 +11,7 @@ const OperationSetting = () => {
     QuotaRemindThreshold: 0,
     PreConsumedQuota: 0,
     ModelRatio: '',
+    CompletionRatio: '',
     GroupRatio: '',
     TopUpLink: '',
     ChatLink: '',
@@ -34,9 +35,12 @@ const OperationSetting = () => {
     if (success) {
       let newInputs = {};
       data.forEach((item) => {
-        if (item.key === 'ModelRatio' || item.key === 'GroupRatio') {
+        if (item.key === 'ModelRatio' || item.key === 'GroupRatio' || item.key === 'CompletionRatio') {
           item.value = JSON.stringify(JSON.parse(item.value), null, 2);
         }
+        if (item.value === '{}') {
+          item.value = '';
+        }
         newInputs[item.key] = item.value;
       });
       setInputs(newInputs);
@@ -101,6 +105,13 @@ const OperationSetting = () => {
           }
           await updateOption('GroupRatio', inputs.GroupRatio);
         }
+        if (originInputs['CompletionRatio'] !== inputs.CompletionRatio) {
+          if (!verifyJSON(inputs.CompletionRatio)) {
+            showError('补全倍率不是合法的 JSON 字符串');
+            return;
+          }
+          await updateOption('CompletionRatio', inputs.CompletionRatio);
+        }
         break;
       case 'quota':
         if (originInputs['QuotaForNewUser'] !== inputs.QuotaForNewUser) {
@@ -271,10 +282,10 @@ const OperationSetting = () => {
               onChange={handleInputChange}
             />
             <Form.Checkbox
-                checked={inputs.AutomaticEnableChannelEnabled === 'true'}
-                label='成功时自动启用通道'
-                name='AutomaticEnableChannelEnabled'
-                onChange={handleInputChange}
+              checked={inputs.AutomaticEnableChannelEnabled === 'true'}
+              label='成功时自动启用通道'
+              name='AutomaticEnableChannelEnabled'
+              onChange={handleInputChange}
             />
           </Form.Group>
           <Form.Button onClick={() => {
@@ -344,6 +355,17 @@ const OperationSetting = () => {
               placeholder='为一个 JSON 文本，键为模型名称，值为倍率'
             />
           </Form.Group>
+          <Form.Group widths='equal'>
+            <Form.TextArea
+              label='补全倍率'
+              name='CompletionRatio'
+              onChange={handleInputChange}
+              style={{ minHeight: 250, fontFamily: 'JetBrains Mono, Consolas' }}
+              autoComplete='new-password'
+              value={inputs.CompletionRatio}
+              placeholder='为一个 JSON 文本，键为模型名称，值为倍率，此处的倍率设置是模型补全倍率相较于提示倍率的比例，使用该设置可强制覆盖 One API 的内部比例'
+            />
+          </Form.Group>
           <Form.Group widths='equal'>
             <Form.TextArea
               label='分组倍率'