feat: Implement cache token ratio for more precise token pricing

2025-03-08 01:30:50 +08:00
parent 81137e0533
commit 4f194f4e6a
18 changed files with 258 additions and 71 deletions
--- a/controller/channel-test.go
+++ b/controller/channel-test.go
@@ -158,7 +158,7 @@ func testChannel(channel *model.Channel, testModel string) (err error, openAIErr
 	tok := time.Now()
 	milliseconds := tok.Sub(tik).Milliseconds()
 	consumedTime := float64(milliseconds) / 1000.0
-	other := service.GenerateTextOtherInfo(c, info, priceData.ModelRatio, priceData.GroupRatio, priceData.CompletionRatio, priceData.ModelPrice)
+	other := service.GenerateTextOtherInfo(c, info, priceData.ModelRatio, priceData.GroupRatio, priceData.CompletionRatio, 0, 0.0, priceData.ModelPrice)
 	model.RecordConsumeLog(c, 1, channel.Id, usage.PromptTokens, usage.CompletionTokens, info.OriginModelName, "模型测试",
 		quota, "模型测试", 0, quota, int(consumedTime), false, info.Group, other)
 	common.SysLog(fmt.Sprintf("testing channel #%d, response: \n%s", channel.Id, string(respBody)))
--- a/controller/pricing.go
+++ b/controller/pricing.go
@@ -4,6 +4,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"one-api/model"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 )

 func GetPricing(c *gin.Context) {
@@ -39,7 +40,7 @@ func GetPricing(c *gin.Context) {
 }

 func ResetModelRatio(c *gin.Context) {
-	defaultStr := setting.DefaultModelRatio2JSONString()
+	defaultStr := operation_setting.DefaultModelRatio2JSONString()
 	err := model.UpdateOption("ModelRatio", defaultStr)
 	if err != nil {
 		c.JSON(200, gin.H{
@@ -48,7 +49,7 @@ func ResetModelRatio(c *gin.Context) {
 		})
 		return
 	}
-	err = setting.UpdateModelRatioByJSONString(defaultStr)
+	err = operation_setting.UpdateModelRatioByJSONString(defaultStr)
 	if err != nil {
 		c.JSON(200, gin.H{
 			"success": false,
--- a/model/option.go
+++ b/model/option.go
@@ -92,11 +92,12 @@ func InitOptionMap() {
 	common.OptionMap["ModelRequestRateLimitCount"] = strconv.Itoa(setting.ModelRequestRateLimitCount)
 	common.OptionMap["ModelRequestRateLimitDurationMinutes"] = strconv.Itoa(setting.ModelRequestRateLimitDurationMinutes)
 	common.OptionMap["ModelRequestRateLimitSuccessCount"] = strconv.Itoa(setting.ModelRequestRateLimitSuccessCount)
-	common.OptionMap["ModelRatio"] = setting.ModelRatio2JSONString()
-	common.OptionMap["ModelPrice"] = setting.ModelPrice2JSONString()
+	common.OptionMap["ModelRatio"] = operation_setting.ModelRatio2JSONString()
+	common.OptionMap["ModelPrice"] = operation_setting.ModelPrice2JSONString()
+	common.OptionMap["CacheRatio"] = operation_setting.CacheRatio2JSONString()
 	common.OptionMap["GroupRatio"] = setting.GroupRatio2JSONString()
 	common.OptionMap["UserUsableGroups"] = setting.UserUsableGroups2JSONString()
-	common.OptionMap["CompletionRatio"] = setting.CompletionRatio2JSONString()
+	common.OptionMap["CompletionRatio"] = operation_setting.CompletionRatio2JSONString()
 	common.OptionMap["TopUpLink"] = common.TopUpLink
 	common.OptionMap["ChatLink"] = common.ChatLink
 	common.OptionMap["ChatLink2"] = common.ChatLink2
@@ -344,15 +345,17 @@ func updateOptionMap(key string, value string) (err error) {
 	case "DataExportDefaultTime":
 		common.DataExportDefaultTime = value
 	case "ModelRatio":
-		err = setting.UpdateModelRatioByJSONString(value)
+		err = operation_setting.UpdateModelRatioByJSONString(value)
 	case "GroupRatio":
 		err = setting.UpdateGroupRatioByJSONString(value)
 	case "UserUsableGroups":
 		err = setting.UpdateUserUsableGroupsByJSONString(value)
 	case "CompletionRatio":
-		err = setting.UpdateCompletionRatioByJSONString(value)
+		err = operation_setting.UpdateCompletionRatioByJSONString(value)
 	case "ModelPrice":
-		err = setting.UpdateModelPriceByJSONString(value)
+		err = operation_setting.UpdateModelPriceByJSONString(value)
+	case "CacheRatio":
+		err = operation_setting.UpdateCacheRatioByJSONString(value)
 	case "TopUpLink":
 		common.TopUpLink = value
 	case "ChatLink":
--- a/model/pricing.go
+++ b/model/pricing.go
@@ -2,7 +2,7 @@ package model

 import (
 	"one-api/common"
-	"one-api/setting"
+	"one-api/setting/operation_setting"
 	"sync"
 	"time"
 )
@@ -65,14 +65,14 @@ func updatePricing() {
 			ModelName:   model,
 			EnableGroup: groups,
 		}
-		modelPrice, findPrice := setting.GetModelPrice(model, false)
+		modelPrice, findPrice := operation_setting.GetModelPrice(model, false)
 		if findPrice {
 			pricing.ModelPrice = modelPrice
 			pricing.QuotaType = 1
 		} else {
-			modelRatio, _ := setting.GetModelRatio(model)
+			modelRatio, _ := operation_setting.GetModelRatio(model)
 			pricing.ModelRatio = modelRatio
-			pricing.CompletionRatio = setting.GetCompletionRatio(model)
+			pricing.CompletionRatio = operation_setting.GetCompletionRatio(model)
 			pricing.QuotaType = 0
 		}
 		pricingMap = append(pricingMap, pricing)
--- a/relay/helper/price.go
+++ b/relay/helper/price.go
@@ -6,30 +6,33 @@ import (
 	"one-api/common"
 	relaycommon "one-api/relay/common"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 )

 type PriceData struct {
 	ModelPrice             float64
 	ModelRatio             float64
 	CompletionRatio        float64
+	CacheRatio             float64
 	GroupRatio             float64
 	UsePrice               bool
 	ShouldPreConsumedQuota int
 }

 func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens int, maxTokens int) (PriceData, error) {
-	modelPrice, usePrice := setting.GetModelPrice(info.OriginModelName, false)
+	modelPrice, usePrice := operation_setting.GetModelPrice(info.OriginModelName, false)
 	groupRatio := setting.GetGroupRatio(info.Group)
 	var preConsumedQuota int
 	var modelRatio float64
 	var completionRatio float64
+	var cacheRatio float64
 	if !usePrice {
 		preConsumedTokens := common.PreConsumedQuota
 		if maxTokens != 0 {
 			preConsumedTokens = promptTokens + maxTokens
 		}
 		var success bool
-		modelRatio, success = setting.GetModelRatio(info.OriginModelName)
+		modelRatio, success = operation_setting.GetModelRatio(info.OriginModelName)
 		if !success {
 			if info.UserId == 1 {
 				return PriceData{}, fmt.Errorf("模型 %s 倍率或价格未配置，请设置或开始自用模式；Model %s ratio or price not set, please set or start self-use mode", info.OriginModelName, info.OriginModelName)
@@ -37,7 +40,8 @@ func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens
 				return PriceData{}, fmt.Errorf("模型 %s 倍率或价格未配置, 请联系管理员设置；Model %s ratio or price not set, please contact administrator to set", info.OriginModelName, info.OriginModelName)
 			}
 		}
-		completionRatio = setting.GetCompletionRatio(info.OriginModelName)
+		completionRatio = operation_setting.GetCompletionRatio(info.OriginModelName)
+		cacheRatio, _ = operation_setting.GetCacheRatio(info.OriginModelName)
 		ratio := modelRatio * groupRatio
 		preConsumedQuota = int(float64(preConsumedTokens) * ratio)
 	} else {
@@ -49,6 +53,7 @@ func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens
 		CompletionRatio:        completionRatio,
 		GroupRatio:             groupRatio,
 		UsePrice:               usePrice,
+		CacheRatio:             cacheRatio,
 		ShouldPreConsumedQuota: preConsumedQuota,
 	}, nil
 }
--- a/relay/relay-mj.go
+++ b/relay/relay-mj.go
@@ -15,6 +15,7 @@ import (
 	relayconstant "one-api/relay/constant"
 	"one-api/service"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 	"strconv"
 	"strings"
 	"time"
@@ -157,10 +158,10 @@ func RelaySwapFace(c *gin.Context) *dto.MidjourneyResponse {
 		return service.MidjourneyErrorWrapper(constant.MjRequestError, "sour_base64_and_target_base64_is_required")
 	}
 	modelName := service.CoverActionToModelName(constant.MjActionSwapFace)
-	modelPrice, success := setting.GetModelPrice(modelName, true)
+	modelPrice, success := operation_setting.GetModelPrice(modelName, true)
 	// 如果没有配置价格，则使用默认价格
 	if !success {
-		defaultPrice, ok := setting.GetDefaultModelRatioMap()[modelName]
+		defaultPrice, ok := operation_setting.GetDefaultModelRatioMap()[modelName]
 		if !ok {
 			modelPrice = 0.1
 		} else {
@@ -463,10 +464,10 @@ func RelayMidjourneySubmit(c *gin.Context, relayMode int) *dto.MidjourneyRespons
 	fullRequestURL := fmt.Sprintf("%s%s", baseURL, requestURL)

 	modelName := service.CoverActionToModelName(midjRequest.Action)
-	modelPrice, success := setting.GetModelPrice(modelName, true)
+	modelPrice, success := operation_setting.GetModelPrice(modelName, true)
 	// 如果没有配置价格，则使用默认价格
 	if !success {
-		defaultPrice, ok := setting.GetDefaultModelRatioMap()[modelName]
+		defaultPrice, ok := operation_setting.GetDefaultModelRatioMap()[modelName]
 		if !ok {
 			modelPrice = 0.1
 		} else {
--- a/relay/relay-text.go
+++ b/relay/relay-text.go
@@ -110,7 +110,7 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
 	if err != nil {
 		return service.OpenAIErrorWrapperLocal(err, "model_price_error", http.StatusInternalServerError)
 	}
-	
+
 	// pre-consume quota 预消耗配额
 	preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
 	if openaiErr != nil {
@@ -304,24 +304,26 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 			CompletionTokens: 0,
 			TotalTokens:      relayInfo.PromptTokens,
 		}
-		extraContent += "  ，（可能是请求出错）"
+		extraContent += "（可能是请求出错）"
 	}
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	promptTokens := usage.PromptTokens
+	cacheTokens := usage.PromptTokensDetails.CachedTokens
 	completionTokens := usage.CompletionTokens
 	modelName := relayInfo.OriginModelName

 	tokenName := ctx.GetString("token_name")
-	completionRatio := setting.GetCompletionRatio(modelName)
+	completionRatio := priceData.CompletionRatio
+	cacheRatio := priceData.CacheRatio
 	ratio := priceData.ModelRatio * priceData.GroupRatio
 	modelRatio := priceData.ModelRatio
 	groupRatio := priceData.GroupRatio
 	modelPrice := priceData.ModelPrice
-	usePrice := priceData.UsePrice

 	quota := 0
 	if !priceData.UsePrice {
-		quota = promptTokens + int(math.Round(float64(completionTokens)*completionRatio))
+		quota = (promptTokens - cacheTokens) + int(math.Round(float64(cacheTokens)*cacheRatio))
+		quota += int(math.Round(float64(completionTokens) * completionRatio))
 		quota = int(math.Round(float64(quota) * ratio))
 		if ratio != 0 && quota <= 0 {
 			quota = 1
@@ -330,8 +332,9 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 		quota = int(modelPrice * common.QuotaPerUnit * groupRatio)
 	}
 	totalTokens := promptTokens + completionTokens
+	
 	var logContent string
-	if !usePrice {
+	if !priceData.UsePrice {
 		logContent = fmt.Sprintf("模型倍率 %.2f，补全倍率 %.2f，分组倍率 %.2f", modelRatio, completionRatio, groupRatio)
 	} else {
 		logContent = fmt.Sprintf("模型价格 %.2f，分组倍率 %.2f", modelPrice, groupRatio)
@@ -372,7 +375,7 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	if extraContent != "" {
 		logContent += ", " + extraContent
 	}
-	other := service.GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, modelPrice)
+	other := service.GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, cacheTokens, cacheRatio, modelPrice)
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, promptTokens, completionTokens, logModel,
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)

--- a/relay/relay_task.go
+++ b/relay/relay_task.go
@@ -16,6 +16,7 @@ import (
 	relayconstant "one-api/relay/constant"
 	"one-api/service"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 )

 /*
@@ -37,9 +38,9 @@ func RelayTaskSubmit(c *gin.Context, relayMode int) (taskErr *dto.TaskError) {
 	}

 	modelName := service.CoverTaskActionToModelName(platform, relayInfo.Action)
-	modelPrice, success := setting.GetModelPrice(modelName, true)
+	modelPrice, success := operation_setting.GetModelPrice(modelName, true)
 	if !success {
-		defaultPrice, ok := setting.GetDefaultModelRatioMap()[modelName]
+		defaultPrice, ok := operation_setting.GetDefaultModelRatioMap()[modelName]
 		if !ok {
 			modelPrice = 0.1
 		} else {
--- a/relay/websocket.go
+++ b/relay/websocket.go
@@ -11,6 +11,7 @@ import (
 	relaycommon "one-api/relay/common"
 	"one-api/service"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 )

 func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWithStatusCode) {
@@ -39,7 +40,7 @@ func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWi
 		}
 	}
 	//relayInfo.UpstreamModelName = textRequest.Model
-	modelPrice, getModelPriceSuccess := setting.GetModelPrice(relayInfo.UpstreamModelName, false)
+	modelPrice, getModelPriceSuccess := operation_setting.GetModelPrice(relayInfo.UpstreamModelName, false)
 	groupRatio := setting.GetGroupRatio(relayInfo.Group)

 	var preConsumedQuota int
@@ -65,7 +66,7 @@ func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWi
 		//if realtimeEvent.Session.MaxResponseOutputTokens != 0 {
 		//	preConsumedTokens = promptTokens + int(realtimeEvent.Session.MaxResponseOutputTokens)
 		//}
-		modelRatio, _ = setting.GetModelRatio(relayInfo.UpstreamModelName)
+		modelRatio, _ = operation_setting.GetModelRatio(relayInfo.UpstreamModelName)
 		ratio = modelRatio * groupRatio
 		preConsumedQuota = int(float64(preConsumedTokens) * ratio)
 	} else {
--- a/service/log_info_generate.go
+++ b/service/log_info_generate.go
@@ -1,16 +1,20 @@
 package service

 import (
-	"github.com/gin-gonic/gin"
 	"one-api/dto"
 	relaycommon "one-api/relay/common"
+
+	"github.com/gin-gonic/gin"
 )

-func GenerateTextOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelRatio, groupRatio, completionRatio, modelPrice float64) map[string]interface{} {
+func GenerateTextOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelRatio, groupRatio, completionRatio float64,
+	cacheTokens int, cacheRatio float64, modelPrice float64) map[string]interface{} {
 	other := make(map[string]interface{})
 	other["model_ratio"] = modelRatio
 	other["group_ratio"] = groupRatio
 	other["completion_ratio"] = completionRatio
+	other["cache_tokens"] = cacheTokens
+	other["cache_ratio"] = cacheRatio
 	other["model_price"] = modelPrice
 	other["frt"] = float64(relayInfo.FirstResponseTime.UnixMilli() - relayInfo.StartTime.UnixMilli())
 	if relayInfo.ReasoningEffort != "" {
@@ -27,7 +31,7 @@ func GenerateTextOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, m
 }

 func GenerateWssOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.RealtimeUsage, modelRatio, groupRatio, completionRatio, audioRatio, audioCompletionRatio, modelPrice float64) map[string]interface{} {
-	info := GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, modelPrice)
+	info := GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, 0, 0.0, modelPrice)
 	info["ws"] = true
 	info["audio_input"] = usage.InputTokenDetails.AudioTokens
 	info["audio_output"] = usage.OutputTokenDetails.AudioTokens
@@ -39,7 +43,7 @@ func GenerateWssOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, us
 }

 func GenerateAudioOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.Usage, modelRatio, groupRatio, completionRatio, audioRatio, audioCompletionRatio, modelPrice float64) map[string]interface{} {
-	info := GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, modelPrice)
+	info := GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, 0, 0.0, modelPrice)
 	info["audio"] = true
 	info["audio_input"] = usage.PromptTokensDetails.AudioTokens
 	info["audio_output"] = usage.CompletionTokenDetails.AudioTokens
--- a/service/quota.go
+++ b/service/quota.go
@@ -12,6 +12,7 @@ import (
 	relaycommon "one-api/relay/common"
 	"one-api/relay/helper"
 	"one-api/setting"
+	"one-api/setting/operation_setting"
 	"strings"
 	"time"

@@ -38,9 +39,9 @@ func calculateAudioQuota(info QuotaInfo) int {
 		return int(info.ModelPrice * common.QuotaPerUnit * info.GroupRatio)
 	}

-	completionRatio := setting.GetCompletionRatio(info.ModelName)
-	audioRatio := setting.GetAudioRatio(info.ModelName)
-	audioCompletionRatio := setting.GetAudioCompletionRatio(info.ModelName)
+	completionRatio := operation_setting.GetCompletionRatio(info.ModelName)
+	audioRatio := operation_setting.GetAudioRatio(info.ModelName)
+	audioCompletionRatio := operation_setting.GetAudioCompletionRatio(info.ModelName)
 	ratio := info.GroupRatio * info.ModelRatio

 	quota := info.InputDetails.TextTokens + int(math.Round(float64(info.OutputDetails.TextTokens)*completionRatio))
@@ -75,7 +76,7 @@ func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usag
 	audioInputTokens := usage.InputTokenDetails.AudioTokens
 	audioOutTokens := usage.OutputTokenDetails.AudioTokens
 	groupRatio := setting.GetGroupRatio(relayInfo.Group)
-	modelRatio, _ := setting.GetModelRatio(modelName)
+	modelRatio, _ := operation_setting.GetModelRatio(modelName)

 	quotaInfo := QuotaInfo{
 		InputDetails: TokenDetails{
@@ -122,9 +123,9 @@ func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, mod
 	audioOutTokens := usage.OutputTokenDetails.AudioTokens

 	tokenName := ctx.GetString("token_name")
-	completionRatio := setting.GetCompletionRatio(modelName)
-	audioRatio := setting.GetAudioRatio(relayInfo.OriginModelName)
-	audioCompletionRatio := setting.GetAudioCompletionRatio(modelName)
+	completionRatio := operation_setting.GetCompletionRatio(modelName)
+	audioRatio := operation_setting.GetAudioRatio(relayInfo.OriginModelName)
+	audioCompletionRatio := operation_setting.GetAudioCompletionRatio(modelName)

 	quotaInfo := QuotaInfo{
 		InputDetails: TokenDetails{
@@ -184,9 +185,9 @@ func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	audioOutTokens := usage.CompletionTokenDetails.AudioTokens

 	tokenName := ctx.GetString("token_name")
-	completionRatio := setting.GetCompletionRatio(relayInfo.OriginModelName)
-	audioRatio := setting.GetAudioRatio(relayInfo.OriginModelName)
-	audioCompletionRatio := setting.GetAudioCompletionRatio(relayInfo.OriginModelName)
+	completionRatio := operation_setting.GetCompletionRatio(relayInfo.OriginModelName)
+	audioRatio := operation_setting.GetAudioRatio(relayInfo.OriginModelName)
+	audioCompletionRatio := operation_setting.GetAudioCompletionRatio(relayInfo.OriginModelName)

 	modelRatio := priceData.ModelRatio
 	groupRatio := priceData.GroupRatio
--- a/service/token_counter.go
+++ b/service/token_counter.go
@@ -10,7 +10,7 @@ import (
 	"one-api/constant"
 	"one-api/dto"
 	relaycommon "one-api/relay/common"
-	"one-api/setting"
+	"one-api/setting/operation_setting"
 	"strings"
 	"unicode/utf8"

@@ -33,7 +33,7 @@ func InitTokenEncoders() {
 	if err != nil {
 		common.FatalLog(fmt.Sprintf("failed to get gpt-4o token encoder: %s", err.Error()))
 	}
-	for model, _ := range setting.GetDefaultModelRatioMap() {
+	for model, _ := range operation_setting.GetDefaultModelRatioMap() {
 		if strings.HasPrefix(model, "gpt-3.5") {
 			tokenEncoderMap[model] = cl100TokenEncoder
 		} else if strings.HasPrefix(model, "gpt-4") {
--- a/setting/operation_setting/cache_ratio.go
+++ b/setting/operation_setting/cache_ratio.go
@@ -0,0 +1,77 @@
+package operation_setting
+
+import (
+	"encoding/json"
+	"one-api/common"
+	"sync"
+)
+
+var defaultCacheRatio = map[string]float64{
+	"gpt-4":                        0.5,
+	"o1-2024-12-17":                0.5,
+	"o1-preview-2024-09-12":        0.5,
+	"o1-mini-2024-09-12":           0.5,
+	"gpt-4o-2024-11-20":            0.5,
+	"gpt-4o-2024-08-06":            0.5,
+	"gpt-4o-mini-2024-07-18":       0.5,
+	"gpt-4o-realtime-preview":      0.5,
+	"gpt-4o-mini-realtime-preview": 0.5,
+	"deepseek-chat":                0.5,
+	"deepseek-reasoner":            0.5,
+	"deepseek-coder":               0.5,
+}
+
+var cacheRatioMap map[string]float64
+var cacheRatioMapMutex sync.RWMutex
+
+// GetCacheRatioMap returns the cache ratio map
+func GetCacheRatioMap() map[string]float64 {
+	cacheRatioMapMutex.Lock()
+	defer cacheRatioMapMutex.Unlock()
+	if cacheRatioMap == nil {
+		cacheRatioMap = defaultCacheRatio
+	}
+	return cacheRatioMap
+}
+
+// CacheRatio2JSONString converts the cache ratio map to a JSON string
+func CacheRatio2JSONString() string {
+	GetCacheRatioMap()
+	jsonBytes, err := json.Marshal(cacheRatioMap)
+	if err != nil {
+		common.SysError("error marshalling cache ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+// UpdateCacheRatioByJSONString updates the cache ratio map from a JSON string
+func UpdateCacheRatioByJSONString(jsonStr string) error {
+	cacheRatioMapMutex.Lock()
+	defer cacheRatioMapMutex.Unlock()
+	cacheRatioMap = make(map[string]float64)
+	return json.Unmarshal([]byte(jsonStr), &cacheRatioMap)
+}
+
+// GetCacheRatio returns the cache ratio for a model
+func GetCacheRatio(name string) (float64, bool) {
+	GetCacheRatioMap()
+	ratio, ok := cacheRatioMap[name]
+	if !ok {
+		return 0.5, false // Default to 0.5 if not found
+	}
+	return ratio, true
+}
+
+// DefaultCacheRatio2JSONString converts the default cache ratio map to a JSON string
+func DefaultCacheRatio2JSONString() string {
+	jsonBytes, err := json.Marshal(defaultCacheRatio)
+	if err != nil {
+		common.SysError("error marshalling default cache ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+// GetDefaultCacheRatioMap returns the default cache ratio map
+func GetDefaultCacheRatioMap() map[string]float64 {
+	return defaultCacheRatio
+}
--- a/setting/operation_setting/model-ratio.go
+++ b/setting/operation_setting/model-ratio.go
@@ -1,9 +1,8 @@
-package setting
+package operation_setting

 import (
 	"encoding/json"
 	"one-api/common"
-	"one-api/setting/operation_setting"
 	"strings"
 	"sync"
 )
@@ -326,7 +325,7 @@ func GetModelRatio(name string) (float64, bool) {
 	}
 	ratio, ok := modelRatioMap[name]
 	if !ok {
-		return 37.5, operation_setting.SelfUseModeEnabled
+		return 37.5, SelfUseModeEnabled
 	}
 	return ratio, true
 }
--- a/web/src/components/LogsTable.js
+++ b/web/src/components/LogsTable.js
@@ -464,6 +464,8 @@ const LogsTable = () => {
          other.model_ratio,
          other.model_price,
          other.group_ratio,
+          other.cache_tokens || 0,
+          other.cache_ratio || 1.0,
        );
        return (
            <Paragraph
@@ -665,6 +667,8 @@ const LogsTable = () => {
            other?.audio_ratio,
            other?.audio_completion_ratio,
            other.group_ratio,
+            other.cache_tokens || 0,
+            other.cache_ratio || 1.0,
          );
        } else {
          content = renderModelPrice(
@@ -674,6 +678,8 @@ const LogsTable = () => {
            other.model_price,
            other.completion_ratio,
            other.group_ratio,
+            other.cache_tokens || 0,
+            other.cache_ratio || 1.0,
          );
        }
        expandDataLocal.push({
--- a/web/src/components/OperationSetting.js
+++ b/web/src/components/OperationSetting.js
@@ -28,6 +28,7 @@ const OperationSetting = () => {
    PreConsumedQuota: 0,
    StreamCacheQueueLength: 0,
    ModelRatio: '',
+    CacheRatio: '',
    CompletionRatio: '',
    ModelPrice: '',
    GroupRatio: '',
@@ -77,7 +78,8 @@ const OperationSetting = () => {
          item.key === 'GroupRatio' ||
          item.key === 'UserUsableGroups' ||
          item.key === 'CompletionRatio' ||
-          item.key === 'ModelPrice'
+          item.key === 'ModelPrice' ||
+          item.key === 'CacheRatio'
        ) {
          item.value = JSON.stringify(JSON.parse(item.value), null, 2);
        }
--- a/web/src/helpers/render.js
+++ b/web/src/helpers/render.js
@@ -298,6 +298,8 @@ export function renderModelPrice(
  modelPrice = -1,
  completionRatio,
  groupRatio,
+  cacheTokens = 0,
+  cacheRatio = 1.0,
 ) {
  if (modelPrice !== -1) {
    return i18next.t('模型价格：${{price}} * 分组倍率：{{ratio}} = ${{total}}', {
@@ -311,9 +313,15 @@ export function renderModelPrice(
    }
    let inputRatioPrice = modelRatio * 2.0;
    let completionRatioPrice = modelRatio * 2.0 * completionRatio;
+    let cacheRatioPrice = modelRatio * 2.0 * cacheRatio;
+    
+    // Calculate effective input tokens (non-cached + cached with ratio applied)
+    const effectiveInputTokens = (inputTokens - cacheTokens) + (cacheTokens * cacheRatio);
+    
    let price =
-      (inputTokens / 1000000) * inputRatioPrice * groupRatio +
+      (effectiveInputTokens / 1000000) * inputRatioPrice * groupRatio +
      (completionTokens / 1000000) * completionRatioPrice * groupRatio;
+    
    return (
      <>
        <article>
@@ -327,16 +335,36 @@ export function renderModelPrice(
            ratio: groupRatio,
            total: completionRatioPrice * groupRatio
          })}</p>
+          {cacheTokens > 0 && (
+            <p>{i18next.t('缓存：${{price}} * {{ratio}} = ${{total}} / 1M tokens (缓存比例: {{cacheRatio}})', {
+              price: cacheRatioPrice,
+              ratio: groupRatio,
+              total: cacheRatioPrice * groupRatio,
+              cacheRatio: cacheRatio
+            })}</p>
+          )}
          <p></p>
          <p>
-            {i18next.t('提示 {{input}} tokens / 1M tokens * ${{price}} + 补全 {{completion}} tokens / 1M tokens * ${{compPrice}} * 分组 {{ratio}} = ${{total}}', {
-              input: inputTokens,
-              price: inputRatioPrice,
-              completion: completionTokens,
-              compPrice: completionRatioPrice,
-              ratio: groupRatio,
-              total: price.toFixed(6)
-            })}
+            {cacheTokens > 0 ? 
+              i18next.t('提示 {{nonCacheInput}} tokens + 缓存 {{cacheInput}} tokens * {{cacheRatio}} / 1M tokens * ${{price}} + 补全 {{completion}} tokens / 1M tokens * ${{compPrice}} * 分组 {{ratio}} = ${{total}}', {
+                nonCacheInput: inputTokens - cacheTokens,
+                cacheInput: cacheTokens,
+                cacheRatio: cacheRatio,
+                price: inputRatioPrice,
+                completion: completionTokens,
+                compPrice: completionRatioPrice,
+                ratio: groupRatio,
+                total: price.toFixed(6)
+              }) :
+              i18next.t('提示 {{input}} tokens / 1M tokens * ${{price}} + 补全 {{completion}} tokens / 1M tokens * ${{compPrice}} * 分组 {{ratio}} = ${{total}}', {
+                input: inputTokens,
+                price: inputRatioPrice,
+                completion: completionTokens,
+                compPrice: completionRatioPrice,
+                ratio: groupRatio,
+                total: price.toFixed(6)
+              })
+            }
          </p>
          <p>{i18next.t('仅供参考，以实际扣费为准')}</p>
        </article>
@@ -349,6 +377,8 @@ export function renderModelPriceSimple(
  modelRatio,
  modelPrice = -1,
  groupRatio,
+  cacheTokens = 0,
+  cacheRatio = 1.0,
 ) {
  if (modelPrice !== -1) {
    return i18next.t('价格：${{price}} * 分组：{{ratio}}', {
@@ -356,10 +386,18 @@ export function renderModelPriceSimple(
      ratio: groupRatio
    });
  } else {
-    return i18next.t('模型: {{ratio}} * 分组: {{groupRatio}}', {
-      ratio: modelRatio,
-      groupRatio: groupRatio
-    });
+    if (cacheTokens !== 0) {
+      return i18next.t('模型: {{ratio}} * 分组: {{groupRatio}} * 缓存比例: {{cacheRatio}}', {
+        ratio: modelRatio,
+        groupRatio: groupRatio,
+        cacheRatio: cacheRatio
+      });
+    } else {
+      return i18next.t('模型: {{ratio}} * 分组: {{groupRatio}}', {
+        ratio: modelRatio,
+        groupRatio: groupRatio
+      });
+    }
  }
 }

@@ -374,6 +412,8 @@ export function renderAudioModelPrice(
  audioRatio,
  audioCompletionRatio,
  groupRatio,
+  cacheTokens = 0,
+  cacheRatio = 1.0,
 ) {
  // 1 ratio = $0.002 / 1K tokens
  if (modelPrice !== -1) {
@@ -388,8 +428,13 @@ export function renderAudioModelPrice(
    // 这里的 *2 是因为 1倍率=0.002刀，请勿删除
    let inputRatioPrice = modelRatio * 2.0;
    let completionRatioPrice = modelRatio * 2.0 * completionRatio;
+    let cacheRatioPrice = modelRatio * 2.0 * cacheRatio;
+    
+    // Calculate effective input tokens (non-cached + cached with ratio applied)
+    const effectiveInputTokens = (inputTokens - cacheTokens) + (cacheTokens * cacheRatio);
+    
    let price =
-      (inputTokens / 1000000) * inputRatioPrice * groupRatio +
+      (effectiveInputTokens / 1000000) * inputRatioPrice * groupRatio +
      (completionTokens / 1000000) * completionRatioPrice * groupRatio +
      (audioInputTokens / 1000000) * inputRatioPrice * audioRatio * groupRatio +
      (audioCompletionTokens / 1000000) * inputRatioPrice * audioRatio * audioCompletionRatio * groupRatio;
@@ -406,6 +451,14 @@ export function renderAudioModelPrice(
            ratio: groupRatio,
            total: completionRatioPrice * groupRatio
          })}</p>
+          {cacheTokens > 0 && (
+            <p>{i18next.t('缓存：${{price}} * {{ratio}} = ${{total}} / 1M tokens (缓存比例: {{cacheRatio}})', {
+              price: cacheRatioPrice,
+              ratio: groupRatio,
+              total: cacheRatioPrice * groupRatio,
+              cacheRatio: cacheRatio
+            })}</p>
+          )}
          <p>{i18next.t('音频提示：${{price}} * {{ratio}} * {{audioRatio}} = ${{total}} / 1M tokens', {
            price: inputRatioPrice,
            ratio: groupRatio,
@@ -420,12 +473,22 @@ export function renderAudioModelPrice(
            total: inputRatioPrice * audioRatio * audioCompletionRatio * groupRatio
          })}</p>
          <p>
-            {i18next.t('文字提示 {{input}} tokens / 1M tokens * ${{price}} + 文字补全 {{completion}} tokens / 1M tokens * ${{compPrice}} +', {
-              input: inputTokens,
-              price: inputRatioPrice,
-              completion: completionTokens,
-              compPrice: completionRatioPrice
-            })}
+            {cacheTokens > 0 ? 
+              i18next.t('文字提示 {{nonCacheInput}} tokens + 文字缓存 {{cacheInput}} tokens * {{cacheRatio}} / 1M tokens * ${{price}} + 文字补全 {{completion}} tokens / 1M tokens * ${{compPrice}} +', {
+                nonCacheInput: inputTokens - cacheTokens,
+                cacheInput: cacheTokens,
+                cacheRatio: cacheRatio,
+                price: inputRatioPrice,
+                completion: completionTokens,
+                compPrice: completionRatioPrice
+              }) :
+              i18next.t('文字提示 {{input}} tokens / 1M tokens * ${{price}} + 文字补全 {{completion}} tokens / 1M tokens * ${{compPrice}} +', {
+                input: inputTokens,
+                price: inputRatioPrice,
+                completion: completionTokens,
+                compPrice: completionRatioPrice
+              })
+            }
          </p>
          <p>
            {i18next.t('音频提示 {{input}} tokens / 1M tokens * ${{price}} * {{audioRatio}} + 音频补全 {{completion}} tokens / 1M tokens * ${{price}} * {{audioRatio}} * {{audioCompRatio}}', {
--- a/web/src/pages/Setting/Operation/ModelRatioSettings.js
+++ b/web/src/pages/Setting/Operation/ModelRatioSettings.js
@@ -15,6 +15,7 @@ export default function ModelRatioSettings(props) {
  const [inputs, setInputs] = useState({
    ModelPrice: '',
    ModelRatio: '',
+    CacheRatio: '',
    CompletionRatio: '',
  });
  const refForm = useRef();
@@ -139,6 +140,25 @@ export default function ModelRatioSettings(props) {
              />
            </Col>
          </Row>
+          <Row gutter={16}>
+            <Col span={16}>
+              <Form.TextArea
+                label={t('提示缓存倍率')}
+                placeholder={t('为一个 JSON 文本，键为模型名称，值为倍率')}
+                field={'CacheRatio'}
+                autosize={{ minRows: 6, maxRows: 12 }}
+                trigger='blur'
+                stopValidateWithError
+                rules={[
+                  {
+                    validator: (rule, value) => verifyJSON(value),
+                    message: '不是合法的 JSON 字符串'
+                  }
+                ]}
+                onChange={(value) => setInputs({ ...inputs, CacheRatio: value })}
+              />
+            </Col>
+          </Row>
          <Row gutter={16}>
            <Col span={16}>
              <Form.TextArea