relay_info.go 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772
  1. package common
  2. import (
  3. "encoding/json"
  4. "errors"
  5. "fmt"
  6. "strings"
  7. "time"
  8. "github.com/QuantumNous/new-api/common"
  9. "github.com/QuantumNous/new-api/constant"
  10. "github.com/QuantumNous/new-api/dto"
  11. relayconstant "github.com/QuantumNous/new-api/relay/constant"
  12. "github.com/QuantumNous/new-api/setting/model_setting"
  13. "github.com/QuantumNous/new-api/types"
  14. "github.com/gin-gonic/gin"
  15. "github.com/gorilla/websocket"
  16. )
  17. type ThinkingContentInfo struct {
  18. IsFirstThinkingContent bool
  19. SendLastThinkingContent bool
  20. HasSentThinkingContent bool
  21. }
  22. const (
  23. LastMessageTypeNone = "none"
  24. LastMessageTypeText = "text"
  25. LastMessageTypeTools = "tools"
  26. LastMessageTypeThinking = "thinking"
  27. )
  28. type ClaudeConvertInfo struct {
  29. LastMessagesType string
  30. Index int
  31. Usage *dto.Usage
  32. FinishReason string
  33. Done bool
  34. ToolCallBaseIndex int
  35. ToolCallMaxIndexOffset int
  36. }
  37. type RerankerInfo struct {
  38. Documents []any
  39. ReturnDocuments bool
  40. }
  41. type BuildInToolInfo struct {
  42. ToolName string
  43. CallCount int
  44. SearchContextSize string
  45. }
  46. type ResponsesUsageInfo struct {
  47. BuiltInTools map[string]*BuildInToolInfo
  48. }
  49. type ChannelMeta struct {
  50. ChannelType int
  51. ChannelId int
  52. ChannelIsMultiKey bool
  53. ChannelMultiKeyIndex int
  54. ChannelBaseUrl string
  55. ApiType int
  56. ApiVersion string
  57. ApiKey string
  58. Organization string
  59. ChannelCreateTime int64
  60. ParamOverride map[string]interface{}
  61. HeadersOverride map[string]interface{}
  62. ChannelSetting dto.ChannelSettings
  63. ChannelOtherSettings dto.ChannelOtherSettings
  64. UpstreamModelName string
  65. IsModelMapped bool
  66. SupportStreamOptions bool // 是否支持流式选项
  67. }
  68. type TokenCountMeta struct {
  69. //promptTokens int
  70. estimatePromptTokens int
  71. }
  72. type RelayInfo struct {
  73. TokenId int
  74. TokenKey string
  75. TokenGroup string
  76. UserId int
  77. UsingGroup string // 使用的分组,当auto跨分组重试时,会变动
  78. UserGroup string // 用户所在分组
  79. TokenUnlimited bool
  80. StartTime time.Time
  81. FirstResponseTime time.Time
  82. isFirstResponse bool
  83. //SendLastReasoningResponse bool
  84. IsStream bool
  85. IsGeminiBatchEmbedding bool
  86. IsPlayground bool
  87. UsePrice bool
  88. RelayMode int
  89. OriginModelName string
  90. RequestURLPath string
  91. ShouldIncludeUsage bool
  92. DisablePing bool // 是否禁止向下游发送自定义 Ping
  93. ClientWs *websocket.Conn
  94. TargetWs *websocket.Conn
  95. InputAudioFormat string
  96. OutputAudioFormat string
  97. RealtimeTools []dto.RealTimeTool
  98. IsFirstRequest bool
  99. AudioUsage bool
  100. ReasoningEffort string
  101. UserSetting dto.UserSetting
  102. UserEmail string
  103. UserQuota int
  104. RelayFormat types.RelayFormat
  105. SendResponseCount int
  106. ReceivedResponseCount int
  107. FinalPreConsumedQuota int // 最终预消耗的配额
  108. // BillingSource indicates whether this request is billed from wallet quota or subscription.
  109. // "" or "wallet" => wallet; "subscription" => subscription
  110. BillingSource string
  111. // SubscriptionId is the user_subscriptions.id used when BillingSource == "subscription"
  112. SubscriptionId int
  113. // SubscriptionPreConsumed is the amount pre-consumed on subscription item (quota units or 1)
  114. SubscriptionPreConsumed int64
  115. // SubscriptionPostDelta is the post-consume delta applied to amount_used (quota units; can be negative).
  116. SubscriptionPostDelta int64
  117. // SubscriptionPlanId / SubscriptionPlanTitle are used for logging/UI display.
  118. SubscriptionPlanId int
  119. SubscriptionPlanTitle string
  120. // RequestId is used for idempotent pre-consume/refund
  121. RequestId string
  122. // SubscriptionAmountTotal / SubscriptionAmountUsedAfterPreConsume are used to compute remaining in logs.
  123. SubscriptionAmountTotal int64
  124. SubscriptionAmountUsedAfterPreConsume int64
  125. IsClaudeBetaQuery bool // /v1/messages?beta=true
  126. IsChannelTest bool // channel test request
  127. PriceData types.PriceData
  128. Request dto.Request
  129. // RequestConversionChain records request format conversions in order, e.g.
  130. // ["openai", "openai_responses"] or ["openai", "claude"].
  131. RequestConversionChain []types.RelayFormat
  132. ThinkingContentInfo
  133. TokenCountMeta
  134. *ClaudeConvertInfo
  135. *RerankerInfo
  136. *ResponsesUsageInfo
  137. *ChannelMeta
  138. *TaskRelayInfo
  139. }
  140. func (info *RelayInfo) InitChannelMeta(c *gin.Context) {
  141. channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  142. paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  143. headerOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelHeaderOverride)
  144. apiType, _ := common.ChannelType2APIType(channelType)
  145. channelMeta := &ChannelMeta{
  146. ChannelType: channelType,
  147. ChannelId: common.GetContextKeyInt(c, constant.ContextKeyChannelId),
  148. ChannelIsMultiKey: common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey),
  149. ChannelMultiKeyIndex: common.GetContextKeyInt(c, constant.ContextKeyChannelMultiKeyIndex),
  150. ChannelBaseUrl: common.GetContextKeyString(c, constant.ContextKeyChannelBaseUrl),
  151. ApiType: apiType,
  152. ApiVersion: c.GetString("api_version"),
  153. ApiKey: common.GetContextKeyString(c, constant.ContextKeyChannelKey),
  154. Organization: c.GetString("channel_organization"),
  155. ChannelCreateTime: c.GetInt64("channel_create_time"),
  156. ParamOverride: paramOverride,
  157. HeadersOverride: headerOverride,
  158. UpstreamModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  159. IsModelMapped: false,
  160. SupportStreamOptions: false,
  161. }
  162. if channelType == constant.ChannelTypeAzure {
  163. channelMeta.ApiVersion = GetAPIVersion(c)
  164. }
  165. if channelType == constant.ChannelTypeVertexAi {
  166. channelMeta.ApiVersion = c.GetString("region")
  167. }
  168. channelSetting, ok := common.GetContextKeyType[dto.ChannelSettings](c, constant.ContextKeyChannelSetting)
  169. if ok {
  170. channelMeta.ChannelSetting = channelSetting
  171. }
  172. channelOtherSettings, ok := common.GetContextKeyType[dto.ChannelOtherSettings](c, constant.ContextKeyChannelOtherSetting)
  173. if ok {
  174. channelMeta.ChannelOtherSettings = channelOtherSettings
  175. }
  176. if streamSupportedChannels[channelMeta.ChannelType] {
  177. channelMeta.SupportStreamOptions = true
  178. }
  179. info.ChannelMeta = channelMeta
  180. // reset some fields based on channel meta
  181. // 重置某些字段,例如模型名称等
  182. if info.Request != nil {
  183. info.Request.SetModelName(info.OriginModelName)
  184. }
  185. }
  186. func (info *RelayInfo) ToString() string {
  187. if info == nil {
  188. return "RelayInfo<nil>"
  189. }
  190. // Basic info
  191. b := &strings.Builder{}
  192. fmt.Fprintf(b, "RelayInfo{ ")
  193. fmt.Fprintf(b, "RelayFormat: %s, ", info.RelayFormat)
  194. fmt.Fprintf(b, "RelayMode: %d, ", info.RelayMode)
  195. fmt.Fprintf(b, "IsStream: %t, ", info.IsStream)
  196. fmt.Fprintf(b, "IsPlayground: %t, ", info.IsPlayground)
  197. fmt.Fprintf(b, "RequestURLPath: %q, ", info.RequestURLPath)
  198. fmt.Fprintf(b, "OriginModelName: %q, ", info.OriginModelName)
  199. fmt.Fprintf(b, "EstimatePromptTokens: %d, ", info.estimatePromptTokens)
  200. fmt.Fprintf(b, "ShouldIncludeUsage: %t, ", info.ShouldIncludeUsage)
  201. fmt.Fprintf(b, "DisablePing: %t, ", info.DisablePing)
  202. fmt.Fprintf(b, "SendResponseCount: %d, ", info.SendResponseCount)
  203. fmt.Fprintf(b, "FinalPreConsumedQuota: %d, ", info.FinalPreConsumedQuota)
  204. // User & token info (mask secrets)
  205. fmt.Fprintf(b, "User{ Id: %d, Email: %q, Group: %q, UsingGroup: %q, Quota: %d }, ",
  206. info.UserId, common.MaskEmail(info.UserEmail), info.UserGroup, info.UsingGroup, info.UserQuota)
  207. fmt.Fprintf(b, "Token{ Id: %d, Unlimited: %t, Key: ***masked*** }, ", info.TokenId, info.TokenUnlimited)
  208. // Time info
  209. latencyMs := info.FirstResponseTime.Sub(info.StartTime).Milliseconds()
  210. fmt.Fprintf(b, "Timing{ Start: %s, FirstResponse: %s, LatencyMs: %d }, ",
  211. info.StartTime.Format(time.RFC3339Nano), info.FirstResponseTime.Format(time.RFC3339Nano), latencyMs)
  212. // Audio / realtime
  213. if info.InputAudioFormat != "" || info.OutputAudioFormat != "" || len(info.RealtimeTools) > 0 || info.AudioUsage {
  214. fmt.Fprintf(b, "Realtime{ AudioUsage: %t, InFmt: %q, OutFmt: %q, Tools: %d }, ",
  215. info.AudioUsage, info.InputAudioFormat, info.OutputAudioFormat, len(info.RealtimeTools))
  216. }
  217. // Reasoning
  218. if info.ReasoningEffort != "" {
  219. fmt.Fprintf(b, "ReasoningEffort: %q, ", info.ReasoningEffort)
  220. }
  221. // Price data (non-sensitive)
  222. if info.PriceData.UsePrice {
  223. fmt.Fprintf(b, "PriceData{ %s }, ", info.PriceData.ToSetting())
  224. }
  225. // Channel metadata (mask ApiKey)
  226. if info.ChannelMeta != nil {
  227. cm := info.ChannelMeta
  228. fmt.Fprintf(b, "ChannelMeta{ Type: %d, Id: %d, IsMultiKey: %t, MultiKeyIndex: %d, BaseURL: %q, ApiType: %d, ApiVersion: %q, Organization: %q, CreateTime: %d, UpstreamModelName: %q, IsModelMapped: %t, SupportStreamOptions: %t, ApiKey: ***masked*** }, ",
  229. cm.ChannelType, cm.ChannelId, cm.ChannelIsMultiKey, cm.ChannelMultiKeyIndex, cm.ChannelBaseUrl, cm.ApiType, cm.ApiVersion, cm.Organization, cm.ChannelCreateTime, cm.UpstreamModelName, cm.IsModelMapped, cm.SupportStreamOptions)
  230. }
  231. // Responses usage info (non-sensitive)
  232. if info.ResponsesUsageInfo != nil && len(info.ResponsesUsageInfo.BuiltInTools) > 0 {
  233. fmt.Fprintf(b, "ResponsesTools{ ")
  234. first := true
  235. for name, tool := range info.ResponsesUsageInfo.BuiltInTools {
  236. if !first {
  237. fmt.Fprintf(b, ", ")
  238. }
  239. first = false
  240. if tool != nil {
  241. fmt.Fprintf(b, "%s: calls=%d", name, tool.CallCount)
  242. } else {
  243. fmt.Fprintf(b, "%s: calls=0", name)
  244. }
  245. }
  246. fmt.Fprintf(b, " }, ")
  247. }
  248. fmt.Fprintf(b, "}")
  249. return b.String()
  250. }
  251. // 定义支持流式选项的通道类型
  252. var streamSupportedChannels = map[int]bool{
  253. constant.ChannelTypeOpenAI: true,
  254. constant.ChannelTypeAnthropic: true,
  255. constant.ChannelTypeAws: true,
  256. constant.ChannelTypeGemini: true,
  257. constant.ChannelCloudflare: true,
  258. constant.ChannelTypeAzure: true,
  259. constant.ChannelTypeVolcEngine: true,
  260. constant.ChannelTypeOllama: true,
  261. constant.ChannelTypeXai: true,
  262. constant.ChannelTypeDeepSeek: true,
  263. constant.ChannelTypeBaiduV2: true,
  264. constant.ChannelTypeZhipu_v4: true,
  265. constant.ChannelTypeAli: true,
  266. constant.ChannelTypeSubmodel: true,
  267. constant.ChannelTypeCodex: true,
  268. }
  269. func GenRelayInfoWs(c *gin.Context, ws *websocket.Conn) *RelayInfo {
  270. info := genBaseRelayInfo(c, nil)
  271. info.RelayFormat = types.RelayFormatOpenAIRealtime
  272. info.ClientWs = ws
  273. info.InputAudioFormat = "pcm16"
  274. info.OutputAudioFormat = "pcm16"
  275. info.IsFirstRequest = true
  276. return info
  277. }
  278. func GenRelayInfoClaude(c *gin.Context, request dto.Request) *RelayInfo {
  279. info := genBaseRelayInfo(c, request)
  280. info.RelayFormat = types.RelayFormatClaude
  281. info.ShouldIncludeUsage = false
  282. info.ClaudeConvertInfo = &ClaudeConvertInfo{
  283. LastMessagesType: LastMessageTypeNone,
  284. }
  285. if c.Query("beta") == "true" {
  286. info.IsClaudeBetaQuery = true
  287. }
  288. return info
  289. }
  290. func GenRelayInfoRerank(c *gin.Context, request *dto.RerankRequest) *RelayInfo {
  291. info := genBaseRelayInfo(c, request)
  292. info.RelayMode = relayconstant.RelayModeRerank
  293. info.RelayFormat = types.RelayFormatRerank
  294. info.RerankerInfo = &RerankerInfo{
  295. Documents: request.Documents,
  296. ReturnDocuments: request.GetReturnDocuments(),
  297. }
  298. return info
  299. }
  300. func GenRelayInfoOpenAIAudio(c *gin.Context, request dto.Request) *RelayInfo {
  301. info := genBaseRelayInfo(c, request)
  302. info.RelayFormat = types.RelayFormatOpenAIAudio
  303. return info
  304. }
  305. func GenRelayInfoEmbedding(c *gin.Context, request dto.Request) *RelayInfo {
  306. info := genBaseRelayInfo(c, request)
  307. info.RelayFormat = types.RelayFormatEmbedding
  308. return info
  309. }
  310. func GenRelayInfoResponses(c *gin.Context, request *dto.OpenAIResponsesRequest) *RelayInfo {
  311. info := genBaseRelayInfo(c, request)
  312. info.RelayMode = relayconstant.RelayModeResponses
  313. info.RelayFormat = types.RelayFormatOpenAIResponses
  314. info.ResponsesUsageInfo = &ResponsesUsageInfo{
  315. BuiltInTools: make(map[string]*BuildInToolInfo),
  316. }
  317. if len(request.Tools) > 0 {
  318. for _, tool := range request.GetToolsMap() {
  319. toolType := common.Interface2String(tool["type"])
  320. info.ResponsesUsageInfo.BuiltInTools[toolType] = &BuildInToolInfo{
  321. ToolName: toolType,
  322. CallCount: 0,
  323. }
  324. switch toolType {
  325. case dto.BuildInToolWebSearchPreview:
  326. searchContextSize := common.Interface2String(tool["search_context_size"])
  327. if searchContextSize == "" {
  328. searchContextSize = "medium"
  329. }
  330. info.ResponsesUsageInfo.BuiltInTools[toolType].SearchContextSize = searchContextSize
  331. }
  332. }
  333. }
  334. return info
  335. }
  336. func GenRelayInfoGemini(c *gin.Context, request dto.Request) *RelayInfo {
  337. info := genBaseRelayInfo(c, request)
  338. info.RelayFormat = types.RelayFormatGemini
  339. info.ShouldIncludeUsage = false
  340. return info
  341. }
  342. func GenRelayInfoImage(c *gin.Context, request dto.Request) *RelayInfo {
  343. info := genBaseRelayInfo(c, request)
  344. info.RelayFormat = types.RelayFormatOpenAIImage
  345. return info
  346. }
  347. func GenRelayInfoOpenAI(c *gin.Context, request dto.Request) *RelayInfo {
  348. info := genBaseRelayInfo(c, request)
  349. info.RelayFormat = types.RelayFormatOpenAI
  350. return info
  351. }
  352. func genBaseRelayInfo(c *gin.Context, request dto.Request) *RelayInfo {
  353. //channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  354. //channelId := common.GetContextKeyInt(c, constant.ContextKeyChannelId)
  355. //paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  356. tokenGroup := common.GetContextKeyString(c, constant.ContextKeyTokenGroup)
  357. // 当令牌分组为空时,表示使用用户分组
  358. if tokenGroup == "" {
  359. tokenGroup = common.GetContextKeyString(c, constant.ContextKeyUserGroup)
  360. }
  361. startTime := common.GetContextKeyTime(c, constant.ContextKeyRequestStartTime)
  362. if startTime.IsZero() {
  363. startTime = time.Now()
  364. }
  365. isStream := false
  366. if request != nil {
  367. isStream = request.IsStream(c)
  368. }
  369. // firstResponseTime = time.Now() - 1 second
  370. reqId := common.GetContextKeyString(c, common.RequestIdKey)
  371. if reqId == "" {
  372. reqId = common.GetTimeString() + common.GetRandomString(8)
  373. }
  374. info := &RelayInfo{
  375. Request: request,
  376. RequestId: reqId,
  377. UserId: common.GetContextKeyInt(c, constant.ContextKeyUserId),
  378. UsingGroup: common.GetContextKeyString(c, constant.ContextKeyUsingGroup),
  379. UserGroup: common.GetContextKeyString(c, constant.ContextKeyUserGroup),
  380. UserQuota: common.GetContextKeyInt(c, constant.ContextKeyUserQuota),
  381. UserEmail: common.GetContextKeyString(c, constant.ContextKeyUserEmail),
  382. OriginModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  383. TokenId: common.GetContextKeyInt(c, constant.ContextKeyTokenId),
  384. TokenKey: common.GetContextKeyString(c, constant.ContextKeyTokenKey),
  385. TokenUnlimited: common.GetContextKeyBool(c, constant.ContextKeyTokenUnlimited),
  386. TokenGroup: tokenGroup,
  387. isFirstResponse: true,
  388. RelayMode: relayconstant.Path2RelayMode(c.Request.URL.Path),
  389. RequestURLPath: c.Request.URL.String(),
  390. IsStream: isStream,
  391. StartTime: startTime,
  392. FirstResponseTime: startTime.Add(-time.Second),
  393. ThinkingContentInfo: ThinkingContentInfo{
  394. IsFirstThinkingContent: true,
  395. SendLastThinkingContent: false,
  396. },
  397. TokenCountMeta: TokenCountMeta{
  398. //promptTokens: common.GetContextKeyInt(c, constant.ContextKeyPromptTokens),
  399. estimatePromptTokens: common.GetContextKeyInt(c, constant.ContextKeyEstimatedTokens),
  400. },
  401. }
  402. if info.RelayMode == relayconstant.RelayModeUnknown {
  403. info.RelayMode = c.GetInt("relay_mode")
  404. }
  405. if strings.HasPrefix(c.Request.URL.Path, "/pg") {
  406. info.IsPlayground = true
  407. info.RequestURLPath = strings.TrimPrefix(info.RequestURLPath, "/pg")
  408. info.RequestURLPath = "/v1" + info.RequestURLPath
  409. }
  410. userSetting, ok := common.GetContextKeyType[dto.UserSetting](c, constant.ContextKeyUserSetting)
  411. if ok {
  412. info.UserSetting = userSetting
  413. }
  414. return info
  415. }
  416. func GenRelayInfo(c *gin.Context, relayFormat types.RelayFormat, request dto.Request, ws *websocket.Conn) (*RelayInfo, error) {
  417. var info *RelayInfo
  418. var err error
  419. switch relayFormat {
  420. case types.RelayFormatOpenAI:
  421. info = GenRelayInfoOpenAI(c, request)
  422. case types.RelayFormatOpenAIAudio:
  423. info = GenRelayInfoOpenAIAudio(c, request)
  424. case types.RelayFormatOpenAIImage:
  425. info = GenRelayInfoImage(c, request)
  426. case types.RelayFormatOpenAIRealtime:
  427. info = GenRelayInfoWs(c, ws)
  428. case types.RelayFormatClaude:
  429. info = GenRelayInfoClaude(c, request)
  430. case types.RelayFormatRerank:
  431. if request, ok := request.(*dto.RerankRequest); ok {
  432. info = GenRelayInfoRerank(c, request)
  433. break
  434. }
  435. err = errors.New("request is not a RerankRequest")
  436. case types.RelayFormatGemini:
  437. info = GenRelayInfoGemini(c, request)
  438. case types.RelayFormatEmbedding:
  439. info = GenRelayInfoEmbedding(c, request)
  440. case types.RelayFormatOpenAIResponses:
  441. if request, ok := request.(*dto.OpenAIResponsesRequest); ok {
  442. info = GenRelayInfoResponses(c, request)
  443. break
  444. }
  445. err = errors.New("request is not a OpenAIResponsesRequest")
  446. case types.RelayFormatOpenAIResponsesCompaction:
  447. if request, ok := request.(*dto.OpenAIResponsesCompactionRequest); ok {
  448. return GenRelayInfoResponsesCompaction(c, request), nil
  449. }
  450. return nil, errors.New("request is not a OpenAIResponsesCompactionRequest")
  451. case types.RelayFormatTask:
  452. info = genBaseRelayInfo(c, nil)
  453. case types.RelayFormatMjProxy:
  454. info = genBaseRelayInfo(c, nil)
  455. default:
  456. err = errors.New("invalid relay format")
  457. }
  458. if err != nil {
  459. return nil, err
  460. }
  461. if info == nil {
  462. return nil, errors.New("failed to build relay info")
  463. }
  464. info.InitRequestConversionChain()
  465. return info, nil
  466. }
  467. func (info *RelayInfo) InitRequestConversionChain() {
  468. if info == nil {
  469. return
  470. }
  471. if len(info.RequestConversionChain) > 0 {
  472. return
  473. }
  474. if info.RelayFormat == "" {
  475. return
  476. }
  477. info.RequestConversionChain = []types.RelayFormat{info.RelayFormat}
  478. }
  479. func (info *RelayInfo) AppendRequestConversion(format types.RelayFormat) {
  480. if info == nil {
  481. return
  482. }
  483. if format == "" {
  484. return
  485. }
  486. if len(info.RequestConversionChain) == 0 {
  487. info.RequestConversionChain = []types.RelayFormat{format}
  488. return
  489. }
  490. last := info.RequestConversionChain[len(info.RequestConversionChain)-1]
  491. if last == format {
  492. return
  493. }
  494. info.RequestConversionChain = append(info.RequestConversionChain, format)
  495. }
  496. func GenRelayInfoResponsesCompaction(c *gin.Context, request *dto.OpenAIResponsesCompactionRequest) *RelayInfo {
  497. info := genBaseRelayInfo(c, request)
  498. if info.RelayMode == relayconstant.RelayModeUnknown {
  499. info.RelayMode = relayconstant.RelayModeResponsesCompact
  500. }
  501. info.RelayFormat = types.RelayFormatOpenAIResponsesCompaction
  502. return info
  503. }
  504. //func (info *RelayInfo) SetPromptTokens(promptTokens int) {
  505. // info.promptTokens = promptTokens
  506. //}
  507. func (info *RelayInfo) SetEstimatePromptTokens(promptTokens int) {
  508. info.estimatePromptTokens = promptTokens
  509. }
  510. func (info *RelayInfo) GetEstimatePromptTokens() int {
  511. return info.estimatePromptTokens
  512. }
  513. func (info *RelayInfo) SetFirstResponseTime() {
  514. if info.isFirstResponse {
  515. info.FirstResponseTime = time.Now()
  516. info.isFirstResponse = false
  517. }
  518. }
  519. func (info *RelayInfo) HasSendResponse() bool {
  520. return info.FirstResponseTime.After(info.StartTime)
  521. }
  522. type TaskRelayInfo struct {
  523. Action string
  524. OriginTaskID string
  525. ConsumeQuota bool
  526. }
  527. type TaskSubmitReq struct {
  528. Prompt string `json:"prompt"`
  529. Model string `json:"model,omitempty"`
  530. Mode string `json:"mode,omitempty"`
  531. Image string `json:"image,omitempty"`
  532. Images []string `json:"images,omitempty"`
  533. Size string `json:"size,omitempty"`
  534. Duration int `json:"duration,omitempty"`
  535. Seconds string `json:"seconds,omitempty"`
  536. InputReference string `json:"input_reference,omitempty"`
  537. Metadata map[string]interface{} `json:"metadata,omitempty"`
  538. }
  539. func (t *TaskSubmitReq) GetPrompt() string {
  540. return t.Prompt
  541. }
  542. func (t *TaskSubmitReq) HasImage() bool {
  543. return len(t.Images) > 0
  544. }
  545. func (t *TaskSubmitReq) UnmarshalJSON(data []byte) error {
  546. type Alias TaskSubmitReq
  547. aux := &struct {
  548. Metadata json.RawMessage `json:"metadata,omitempty"`
  549. *Alias
  550. }{
  551. Alias: (*Alias)(t),
  552. }
  553. if err := common.Unmarshal(data, &aux); err != nil {
  554. return err
  555. }
  556. if len(aux.Metadata) > 0 {
  557. var metadataStr string
  558. if err := common.Unmarshal(aux.Metadata, &metadataStr); err == nil && metadataStr != "" {
  559. var metadataObj map[string]interface{}
  560. if err := common.Unmarshal([]byte(metadataStr), &metadataObj); err == nil {
  561. t.Metadata = metadataObj
  562. return nil
  563. }
  564. }
  565. var metadataObj map[string]interface{}
  566. if err := common.Unmarshal(aux.Metadata, &metadataObj); err == nil {
  567. t.Metadata = metadataObj
  568. }
  569. }
  570. return nil
  571. }
  572. func (t *TaskSubmitReq) UnmarshalMetadata(v any) error {
  573. metadata := t.Metadata
  574. if metadata != nil {
  575. metadataBytes, err := json.Marshal(metadata)
  576. if err != nil {
  577. return fmt.Errorf("marshal metadata failed: %w", err)
  578. }
  579. err = json.Unmarshal(metadataBytes, v)
  580. if err != nil {
  581. return fmt.Errorf("unmarshal metadata to target failed: %w", err)
  582. }
  583. }
  584. return nil
  585. }
  586. type TaskInfo struct {
  587. Code int `json:"code"`
  588. TaskID string `json:"task_id"`
  589. Status string `json:"status"`
  590. Reason string `json:"reason,omitempty"`
  591. Url string `json:"url,omitempty"`
  592. RemoteUrl string `json:"remote_url,omitempty"`
  593. Progress string `json:"progress,omitempty"`
  594. CompletionTokens int `json:"completion_tokens,omitempty"` // 用于按倍率计费
  595. TotalTokens int `json:"total_tokens,omitempty"` // 用于按倍率计费
  596. }
  597. func FailTaskInfo(reason string) *TaskInfo {
  598. return &TaskInfo{
  599. Status: "FAILURE",
  600. Reason: reason,
  601. }
  602. }
  603. // RemoveDisabledFields 从请求 JSON 数据中移除渠道设置中禁用的字段
  604. // service_tier: 服务层级字段,可能导致额外计费(OpenAI、Claude、Responses API 支持)
  605. // store: 数据存储授权字段,涉及用户隐私(仅 OpenAI、Responses API 支持,默认允许透传,禁用后可能导致 Codex 无法使用)
  606. // safety_identifier: 安全标识符,用于向 OpenAI 报告违规用户(仅 OpenAI 支持,涉及用户隐私)
  607. func RemoveDisabledFields(jsonData []byte, channelOtherSettings dto.ChannelOtherSettings) ([]byte, error) {
  608. var data map[string]interface{}
  609. if err := common.Unmarshal(jsonData, &data); err != nil {
  610. common.SysError("RemoveDisabledFields Unmarshal error :" + err.Error())
  611. return jsonData, nil
  612. }
  613. // 默认移除 service_tier,除非明确允许(避免额外计费风险)
  614. if !channelOtherSettings.AllowServiceTier {
  615. if _, exists := data["service_tier"]; exists {
  616. delete(data, "service_tier")
  617. }
  618. }
  619. // 默认允许 store 透传,除非明确禁用(禁用可能影响 Codex 使用)
  620. if channelOtherSettings.DisableStore {
  621. if _, exists := data["store"]; exists {
  622. delete(data, "store")
  623. }
  624. }
  625. // 默认移除 safety_identifier,除非明确允许(保护用户隐私,避免向 OpenAI 报告用户信息)
  626. if !channelOtherSettings.AllowSafetyIdentifier {
  627. if _, exists := data["safety_identifier"]; exists {
  628. delete(data, "safety_identifier")
  629. }
  630. }
  631. jsonDataAfter, err := common.Marshal(data)
  632. if err != nil {
  633. common.SysError("RemoveDisabledFields Marshal error :" + err.Error())
  634. return jsonData, nil
  635. }
  636. return jsonDataAfter, nil
  637. }
  638. // RemoveGeminiDisabledFields removes disabled fields from Gemini request JSON data
  639. // Currently supports removing functionResponse.id field which Vertex AI does not support
  640. func RemoveGeminiDisabledFields(jsonData []byte) ([]byte, error) {
  641. if !model_setting.GetGeminiSettings().RemoveFunctionResponseIdEnabled {
  642. return jsonData, nil
  643. }
  644. var data map[string]interface{}
  645. if err := common.Unmarshal(jsonData, &data); err != nil {
  646. common.SysError("RemoveGeminiDisabledFields Unmarshal error: " + err.Error())
  647. return jsonData, nil
  648. }
  649. // Process contents array
  650. // Handle both camelCase (functionResponse) and snake_case (function_response)
  651. if contents, ok := data["contents"].([]interface{}); ok {
  652. for _, content := range contents {
  653. if contentMap, ok := content.(map[string]interface{}); ok {
  654. if parts, ok := contentMap["parts"].([]interface{}); ok {
  655. for _, part := range parts {
  656. if partMap, ok := part.(map[string]interface{}); ok {
  657. // Check functionResponse (camelCase)
  658. if funcResp, ok := partMap["functionResponse"].(map[string]interface{}); ok {
  659. delete(funcResp, "id")
  660. }
  661. // Check function_response (snake_case)
  662. if funcResp, ok := partMap["function_response"].(map[string]interface{}); ok {
  663. delete(funcResp, "id")
  664. }
  665. }
  666. }
  667. }
  668. }
  669. }
  670. }
  671. jsonDataAfter, err := common.Marshal(data)
  672. if err != nil {
  673. common.SysError("RemoveGeminiDisabledFields Marshal error: " + err.Error())
  674. return jsonData, nil
  675. }
  676. return jsonDataAfter, nil
  677. }