Merge branch 'main' into patch/gpt-4o-audio

relay/adaptor/openai/token.go

@@ -10,6 +10,7 @@ import (
 
 	"github.com/pkg/errors"
 	"github.com/pkoukk/tiktoken-go"
+
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/image"
@@ -27,7 +28,8 @@ func InitTokenEncoders() {
 	logger.SysLog("initializing token encoders")
 	gpt35TokenEncoder, err := tiktoken.EncodingForModel("gpt-3.5-turbo")
 	if err != nil {
-		logger.FatalLog(fmt.Sprintf("failed to get gpt-3.5-turbo token encoder: %s", err.Error()))
+		logger.FatalLog(fmt.Sprintf("failed to get gpt-3.5-turbo token encoder: %s, "+
+			"if you are using in offline environment, please set TIKTOKEN_CACHE_DIR to use exsited files, check this link for more information: https://stackoverflow.com/questions/76106366/how-to-use-tiktoken-in-offline-mode-computer ", err.Error()))
 	}
 	defaultTokenEncoder = gpt35TokenEncoder
 	gpt4oTokenEncoder, err := tiktoken.EncodingForModel("gpt-4o")

relay/billing/billing.go

@@ -3,6 +3,7 @@ package billing
 import (
 	"context"
 	"fmt"
+
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/model"
 )
@@ -31,8 +32,17 @@ func PostConsumeQuota(ctx context.Context, tokenId int, quotaDelta int64, totalQ
 	}
 	// totalQuota is total quota consumed
 	if totalQuota != 0 {
-		logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
-		model.RecordConsumeLog(ctx, userId, channelId, int(totalQuota), 0, modelName, tokenName, totalQuota, logContent)
+		logContent := fmt.Sprintf("倍率：%.2f × %.2f", modelRatio, groupRatio)
+		model.RecordConsumeLog(ctx, &model.Log{
+			UserId:           userId,
+			ChannelId:        channelId,
+			PromptTokens:     int(totalQuota),
+			CompletionTokens: 0,
+			ModelName:        modelName,
+			TokenName:        tokenName,
+			Quota:            int(totalQuota),
+			Content:          logContent,
+		})
 		model.UpdateUserUsedQuotaAndRequestCount(userId, totalQuota)
 		model.UpdateChannelUsedQuota(channelId, totalQuota)
 	}

relay/controller/helper.go

@@ -8,7 +8,11 @@ import (
 	"net/http"
 	"strings"
 
+	"github.com/songquanpeng/one-api/common/helper"
+	"github.com/songquanpeng/one-api/relay/constant/role"
+
 	"github.com/gin-gonic/gin"
+
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/logger"
@@ -119,12 +123,20 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
 	if err != nil {
 		logger.Error(ctx, "error update user quota cache: "+err.Error())
 	}
-	var extraLog string
-	if systemPromptReset {
-		extraLog = " （注意系统提示词已被重置）"
-	}
-	logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f%s", modelRatio, groupRatio, completionRatio, extraLog)
-	model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
+	logContent := fmt.Sprintf("倍率：%.2f × %.2f × %.2f", modelRatio, groupRatio, completionRatio)
+	model.RecordConsumeLog(ctx, &model.Log{
+		UserId:            meta.UserId,
+		ChannelId:         meta.ChannelId,
+		PromptTokens:      promptTokens,
+		CompletionTokens:  completionTokens,
+		ModelName:         textRequest.Model,
+		TokenName:         meta.TokenName,
+		Quota:             int(quota),
+		Content:           logContent,
+		IsStream:          meta.IsStream,
+		ElapsedTime:       helper.CalcElapsedTime(meta.StartTime),
+		SystemPromptReset: systemPromptReset,
+	})
 	model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
 	model.UpdateChannelUsedQuota(meta.ChannelId, quota)
 }

relay/controller/image.go

@@ -211,8 +211,17 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		}
 		if quota != 0 {
 			tokenName := c.GetString(ctxkey.TokenName)
-			logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
-			model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, 0, 0, imageRequest.Model, tokenName, quota, logContent)
+			logContent := fmt.Sprintf("倍率：%.2f × %.2f", modelRatio, groupRatio)
+			model.RecordConsumeLog(ctx, &model.Log{
+				UserId:           meta.UserId,
+				ChannelId:        meta.ChannelId,
+				PromptTokens:     0,
+				CompletionTokens: 0,
+				ModelName:        imageRequest.Model,
+				TokenName:        tokenName,
+				Quota:            int(quota),
+				Content:          logContent,
+			})
 			model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
 			channelId := c.GetInt(ctxkey.ChannelId)
 			model.UpdateChannelUsedQuota(channelId, quota)

relay/meta/relay_meta.go

@@ -2,8 +2,10 @@ package meta
 
 import (
 	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
+
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/channeltype"
@@ -32,6 +34,7 @@ type Meta struct {
 	RequestURLPath  string
 	PromptTokens    int // only for DoResponse
 	SystemPrompt    string
+	StartTime       time.Time
 }
 
 // GetMappedModelName returns the mapped model name and a bool indicating if the model name is mapped
@@ -64,6 +67,7 @@ func GetByContext(c *gin.Context) *Meta {
 		APIKey:          strings.TrimPrefix(c.Request.Header.Get("Authorization"), "Bearer "),
 		RequestURLPath:  c.Request.URL.String(),
 		SystemPrompt:    c.GetString(ctxkey.SystemPrompt),
+		StartTime:       time.Now(),
 	}
 	cfg, ok := c.Get(ctxkey.Config)
 	if ok {