oadmin/new-api
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Implement cache token ratio for more precise token pricing

Browse Source
This commit is contained in:
1808837298@qq.com
2025-03-08 01:30:50 +08:00
parent 81137e0533
commit 4f194f4e6a
18 changed files with 258 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
relay/relay-text.go
View File

@@ -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)