feat: Implement cache token ratio for more precise token pricing

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
 }