quota.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352
  1. package service
  2. import (
  3. "errors"
  4. "fmt"
  5. "one-api/common"
  6. constant2 "one-api/constant"
  7. "one-api/dto"
  8. "one-api/model"
  9. relaycommon "one-api/relay/common"
  10. "one-api/relay/helper"
  11. "one-api/setting"
  12. "one-api/setting/operation_setting"
  13. "strings"
  14. "time"
  15. "github.com/bytedance/gopkg/util/gopool"
  16. "github.com/gin-gonic/gin"
  17. "github.com/shopspring/decimal"
  18. )
  19. type TokenDetails struct {
  20. TextTokens int
  21. AudioTokens int
  22. }
  23. type QuotaInfo struct {
  24. InputDetails TokenDetails
  25. OutputDetails TokenDetails
  26. ModelName string
  27. UsePrice bool
  28. ModelPrice float64
  29. ModelRatio float64
  30. GroupRatio float64
  31. }
  32. func calculateAudioQuota(info QuotaInfo) int {
  33. if info.UsePrice {
  34. modelPrice := decimal.NewFromFloat(info.ModelPrice)
  35. quotaPerUnit := decimal.NewFromFloat(common.QuotaPerUnit)
  36. groupRatio := decimal.NewFromFloat(info.GroupRatio)
  37. quota := modelPrice.Mul(quotaPerUnit).Mul(groupRatio)
  38. return int(quota.IntPart())
  39. }
  40. completionRatio := decimal.NewFromFloat(operation_setting.GetCompletionRatio(info.ModelName))
  41. audioRatio := decimal.NewFromFloat(operation_setting.GetAudioRatio(info.ModelName))
  42. audioCompletionRatio := decimal.NewFromFloat(operation_setting.GetAudioCompletionRatio(info.ModelName))
  43. groupRatio := decimal.NewFromFloat(info.GroupRatio)
  44. modelRatio := decimal.NewFromFloat(info.ModelRatio)
  45. ratio := groupRatio.Mul(modelRatio)
  46. inputTextTokens := decimal.NewFromInt(int64(info.InputDetails.TextTokens))
  47. outputTextTokens := decimal.NewFromInt(int64(info.OutputDetails.TextTokens))
  48. inputAudioTokens := decimal.NewFromInt(int64(info.InputDetails.AudioTokens))
  49. outputAudioTokens := decimal.NewFromInt(int64(info.OutputDetails.AudioTokens))
  50. quota := decimal.Zero
  51. quota = quota.Add(inputTextTokens)
  52. quota = quota.Add(outputTextTokens.Mul(completionRatio))
  53. quota = quota.Add(inputAudioTokens.Mul(audioRatio))
  54. quota = quota.Add(outputAudioTokens.Mul(audioRatio).Mul(audioCompletionRatio))
  55. quota = quota.Mul(ratio)
  56. // If ratio is not zero and quota is less than or equal to zero, set quota to 1
  57. if !ratio.IsZero() && quota.LessThanOrEqual(decimal.Zero) {
  58. quota = decimal.NewFromInt(1)
  59. }
  60. return int(quota.Round(0).IntPart())
  61. }
  62. func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.RealtimeUsage) error {
  63. if relayInfo.UsePrice {
  64. return nil
  65. }
  66. userQuota, err := model.GetUserQuota(relayInfo.UserId, false)
  67. if err != nil {
  68. return err
  69. }
  70. token, err := model.GetTokenByKey(strings.TrimLeft(relayInfo.TokenKey, "sk-"), false)
  71. if err != nil {
  72. return err
  73. }
  74. modelName := relayInfo.OriginModelName
  75. textInputTokens := usage.InputTokenDetails.TextTokens
  76. textOutTokens := usage.OutputTokenDetails.TextTokens
  77. audioInputTokens := usage.InputTokenDetails.AudioTokens
  78. audioOutTokens := usage.OutputTokenDetails.AudioTokens
  79. groupRatio := setting.GetGroupRatio(relayInfo.Group)
  80. modelRatio, _ := operation_setting.GetModelRatio(modelName)
  81. quotaInfo := QuotaInfo{
  82. InputDetails: TokenDetails{
  83. TextTokens: textInputTokens,
  84. AudioTokens: audioInputTokens,
  85. },
  86. OutputDetails: TokenDetails{
  87. TextTokens: textOutTokens,
  88. AudioTokens: audioOutTokens,
  89. },
  90. ModelName: modelName,
  91. UsePrice: relayInfo.UsePrice,
  92. ModelRatio: modelRatio,
  93. GroupRatio: groupRatio,
  94. }
  95. quota := calculateAudioQuota(quotaInfo)
  96. if userQuota < quota {
  97. return fmt.Errorf("user quota is not enough, user quota: %s, need quota: %s", common.FormatQuota(userQuota), common.FormatQuota(quota))
  98. }
  99. if !token.UnlimitedQuota && token.RemainQuota < quota {
  100. return fmt.Errorf("token quota is not enough, token remain quota: %s, need quota: %s", common.FormatQuota(token.RemainQuota), common.FormatQuota(quota))
  101. }
  102. err = PostConsumeQuota(relayInfo, quota, 0, false)
  103. if err != nil {
  104. return err
  105. }
  106. common.LogInfo(ctx, "realtime streaming consume quota success, quota: "+fmt.Sprintf("%d", quota))
  107. return nil
  108. }
  109. func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelName string,
  110. usage *dto.RealtimeUsage, preConsumedQuota int, userQuota int, modelRatio float64, groupRatio float64,
  111. modelPrice float64, usePrice bool, extraContent string) {
  112. useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
  113. textInputTokens := usage.InputTokenDetails.TextTokens
  114. textOutTokens := usage.OutputTokenDetails.TextTokens
  115. audioInputTokens := usage.InputTokenDetails.AudioTokens
  116. audioOutTokens := usage.OutputTokenDetails.AudioTokens
  117. tokenName := ctx.GetString("token_name")
  118. completionRatio := decimal.NewFromFloat(operation_setting.GetCompletionRatio(modelName))
  119. audioRatio := decimal.NewFromFloat(operation_setting.GetAudioRatio(relayInfo.OriginModelName))
  120. audioCompletionRatio := decimal.NewFromFloat(operation_setting.GetAudioCompletionRatio(modelName))
  121. quotaInfo := QuotaInfo{
  122. InputDetails: TokenDetails{
  123. TextTokens: textInputTokens,
  124. AudioTokens: audioInputTokens,
  125. },
  126. OutputDetails: TokenDetails{
  127. TextTokens: textOutTokens,
  128. AudioTokens: audioOutTokens,
  129. },
  130. ModelName: modelName,
  131. UsePrice: usePrice,
  132. ModelRatio: modelRatio,
  133. GroupRatio: groupRatio,
  134. }
  135. quota := calculateAudioQuota(quotaInfo)
  136. totalTokens := usage.TotalTokens
  137. var logContent string
  138. if !usePrice {
  139. logContent = fmt.Sprintf("模型倍率 %.2f,补全倍率 %.2f,音频倍率 %.2f,音频补全倍率 %.2f,分组倍率 %.2f",
  140. modelRatio, completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), groupRatio)
  141. } else {
  142. logContent = fmt.Sprintf("模型价格 %.2f,分组倍率 %.2f", modelPrice, groupRatio)
  143. }
  144. // record all the consume log even if quota is 0
  145. if totalTokens == 0 {
  146. // in this case, must be some error happened
  147. // we cannot just return, because we may have to return the pre-consumed quota
  148. quota = 0
  149. logContent += fmt.Sprintf("(可能是上游超时)")
  150. common.LogError(ctx, fmt.Sprintf("total tokens is 0, cannot consume quota, userId %d, channelId %d, "+
  151. "tokenId %d, model %s, pre-consumed quota %d", relayInfo.UserId, relayInfo.ChannelId, relayInfo.TokenId, modelName, preConsumedQuota))
  152. } else {
  153. model.UpdateUserUsedQuotaAndRequestCount(relayInfo.UserId, quota)
  154. model.UpdateChannelUsedQuota(relayInfo.ChannelId, quota)
  155. }
  156. logModel := modelName
  157. if extraContent != "" {
  158. logContent += ", " + extraContent
  159. }
  160. other := GenerateWssOtherInfo(ctx, relayInfo, usage, modelRatio, groupRatio,
  161. completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice)
  162. model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, usage.InputTokens, usage.OutputTokens, logModel,
  163. tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
  164. }
  165. func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
  166. usage *dto.Usage, preConsumedQuota int, userQuota int, priceData helper.PriceData, extraContent string) {
  167. useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
  168. textInputTokens := usage.PromptTokensDetails.TextTokens
  169. textOutTokens := usage.CompletionTokenDetails.TextTokens
  170. audioInputTokens := usage.PromptTokensDetails.AudioTokens
  171. audioOutTokens := usage.CompletionTokenDetails.AudioTokens
  172. tokenName := ctx.GetString("token_name")
  173. completionRatio := decimal.NewFromFloat(operation_setting.GetCompletionRatio(relayInfo.OriginModelName))
  174. audioRatio := decimal.NewFromFloat(operation_setting.GetAudioRatio(relayInfo.OriginModelName))
  175. audioCompletionRatio := decimal.NewFromFloat(operation_setting.GetAudioCompletionRatio(relayInfo.OriginModelName))
  176. modelRatio := priceData.ModelRatio
  177. groupRatio := priceData.GroupRatio
  178. modelPrice := priceData.ModelPrice
  179. usePrice := priceData.UsePrice
  180. quotaInfo := QuotaInfo{
  181. InputDetails: TokenDetails{
  182. TextTokens: textInputTokens,
  183. AudioTokens: audioInputTokens,
  184. },
  185. OutputDetails: TokenDetails{
  186. TextTokens: textOutTokens,
  187. AudioTokens: audioOutTokens,
  188. },
  189. ModelName: relayInfo.OriginModelName,
  190. UsePrice: usePrice,
  191. ModelRatio: modelRatio,
  192. GroupRatio: groupRatio,
  193. }
  194. quota := calculateAudioQuota(quotaInfo)
  195. totalTokens := usage.TotalTokens
  196. var logContent string
  197. if !usePrice {
  198. logContent = fmt.Sprintf("模型倍率 %.2f,补全倍率 %.2f,音频倍率 %.2f,音频补全倍率 %.2f,分组倍率 %.2f",
  199. modelRatio, completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), groupRatio)
  200. } else {
  201. logContent = fmt.Sprintf("模型价格 %.2f,分组倍率 %.2f", modelPrice, groupRatio)
  202. }
  203. // record all the consume log even if quota is 0
  204. if totalTokens == 0 {
  205. // in this case, must be some error happened
  206. // we cannot just return, because we may have to return the pre-consumed quota
  207. quota = 0
  208. logContent += fmt.Sprintf("(可能是上游超时)")
  209. common.LogError(ctx, fmt.Sprintf("total tokens is 0, cannot consume quota, userId %d, channelId %d, "+
  210. "tokenId %d, model %s, pre-consumed quota %d", relayInfo.UserId, relayInfo.ChannelId, relayInfo.TokenId, relayInfo.OriginModelName, preConsumedQuota))
  211. } else {
  212. quotaDelta := quota - preConsumedQuota
  213. if quotaDelta != 0 {
  214. err := PostConsumeQuota(relayInfo, quotaDelta, preConsumedQuota, true)
  215. if err != nil {
  216. common.LogError(ctx, "error consuming token remain quota: "+err.Error())
  217. }
  218. }
  219. model.UpdateUserUsedQuotaAndRequestCount(relayInfo.UserId, quota)
  220. model.UpdateChannelUsedQuota(relayInfo.ChannelId, quota)
  221. }
  222. logModel := relayInfo.OriginModelName
  223. if extraContent != "" {
  224. logContent += ", " + extraContent
  225. }
  226. other := GenerateAudioOtherInfo(ctx, relayInfo, usage, modelRatio, groupRatio,
  227. completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice)
  228. model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, usage.PromptTokens, usage.CompletionTokens, logModel,
  229. tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
  230. }
  231. func PreConsumeTokenQuota(relayInfo *relaycommon.RelayInfo, quota int) error {
  232. if quota < 0 {
  233. return errors.New("quota 不能为负数!")
  234. }
  235. if relayInfo.IsPlayground {
  236. return nil
  237. }
  238. //if relayInfo.TokenUnlimited {
  239. // return nil
  240. //}
  241. token, err := model.GetTokenByKey(relayInfo.TokenKey, false)
  242. if err != nil {
  243. return err
  244. }
  245. if !relayInfo.TokenUnlimited && token.RemainQuota < quota {
  246. return fmt.Errorf("token quota is not enough, token remain quota: %s, need quota: %s", common.FormatQuota(token.RemainQuota), common.FormatQuota(quota))
  247. }
  248. err = model.DecreaseTokenQuota(relayInfo.TokenId, relayInfo.TokenKey, quota)
  249. if err != nil {
  250. return err
  251. }
  252. return nil
  253. }
  254. func PostConsumeQuota(relayInfo *relaycommon.RelayInfo, quota int, preConsumedQuota int, sendEmail bool) (err error) {
  255. if quota > 0 {
  256. err = model.DecreaseUserQuota(relayInfo.UserId, quota)
  257. } else {
  258. err = model.IncreaseUserQuota(relayInfo.UserId, -quota, false)
  259. }
  260. if err != nil {
  261. return err
  262. }
  263. if !relayInfo.IsPlayground {
  264. if quota > 0 {
  265. err = model.DecreaseTokenQuota(relayInfo.TokenId, relayInfo.TokenKey, quota)
  266. } else {
  267. err = model.IncreaseTokenQuota(relayInfo.TokenId, relayInfo.TokenKey, -quota)
  268. }
  269. if err != nil {
  270. return err
  271. }
  272. }
  273. if sendEmail {
  274. if (quota + preConsumedQuota) != 0 {
  275. checkAndSendQuotaNotify(relayInfo, quota, preConsumedQuota)
  276. }
  277. }
  278. return nil
  279. }
  280. func checkAndSendQuotaNotify(relayInfo *relaycommon.RelayInfo, quota int, preConsumedQuota int) {
  281. gopool.Go(func() {
  282. userSetting := relayInfo.UserSetting
  283. threshold := common.QuotaRemindThreshold
  284. if userCustomThreshold, ok := userSetting[constant2.UserSettingQuotaWarningThreshold]; ok {
  285. threshold = int(userCustomThreshold.(float64))
  286. }
  287. //noMoreQuota := userCache.Quota-(quota+preConsumedQuota) <= 0
  288. quotaTooLow := false
  289. consumeQuota := quota + preConsumedQuota
  290. if relayInfo.UserQuota-consumeQuota < threshold {
  291. quotaTooLow = true
  292. }
  293. if quotaTooLow {
  294. prompt := "您的额度即将用尽"
  295. topUpLink := fmt.Sprintf("%s/topup", setting.ServerAddress)
  296. content := "{{value}},当前剩余额度为 {{value}},为了不影响您的使用,请及时充值。<br/>充值链接:<a href='{{value}}'>{{value}}</a>"
  297. err := NotifyUser(relayInfo.UserId, relayInfo.UserEmail, relayInfo.UserSetting, dto.NewNotify(dto.NotifyTypeQuotaExceed, prompt, content, []interface{}{prompt, common.FormatQuota(relayInfo.UserQuota), topUpLink, topUpLink}))
  298. if err != nil {
  299. common.SysError(fmt.Sprintf("failed to send quota notify to user %d: %s", relayInfo.UserId, err.Error()))
  300. }
  301. }
  302. })
  303. }