relay_info.go 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811
  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. // ForcePreConsume 为 true 时禁用 BillingSession 的信任额度旁路,
  109. // 强制预扣全额。用于异步任务(视频/音乐生成等),因为请求返回后任务仍在运行,
  110. // 必须在提交前锁定全额。
  111. ForcePreConsume bool
  112. // Billing 是计费会话,封装了预扣费/结算/退款的统一生命周期。
  113. // 免费模型时为 nil。
  114. Billing BillingSettler
  115. // BillingSource indicates whether this request is billed from wallet quota or subscription.
  116. // "" or "wallet" => wallet; "subscription" => subscription
  117. BillingSource string
  118. // SubscriptionId is the user_subscriptions.id used when BillingSource == "subscription"
  119. SubscriptionId int
  120. // SubscriptionPreConsumed is the amount pre-consumed on subscription item (quota units or 1)
  121. SubscriptionPreConsumed int64
  122. // SubscriptionPostDelta is the post-consume delta applied to amount_used (quota units; can be negative).
  123. SubscriptionPostDelta int64
  124. // SubscriptionPlanId / SubscriptionPlanTitle are used for logging/UI display.
  125. SubscriptionPlanId int
  126. SubscriptionPlanTitle string
  127. // RequestId is used for idempotent pre-consume/refund
  128. RequestId string
  129. // SubscriptionAmountTotal / SubscriptionAmountUsedAfterPreConsume are used to compute remaining in logs.
  130. SubscriptionAmountTotal int64
  131. SubscriptionAmountUsedAfterPreConsume int64
  132. IsClaudeBetaQuery bool // /v1/messages?beta=true
  133. IsChannelTest bool // channel test request
  134. PriceData types.PriceData
  135. Request dto.Request
  136. // RequestConversionChain records request format conversions in order, e.g.
  137. // ["openai", "openai_responses"] or ["openai", "claude"].
  138. RequestConversionChain []types.RelayFormat
  139. // 最终请求到上游的格式。可由 adaptor 显式设置;
  140. // 若为空,调用 GetFinalRequestRelayFormat 会回退到 RequestConversionChain 的最后一项或 RelayFormat。
  141. FinalRequestRelayFormat types.RelayFormat
  142. ThinkingContentInfo
  143. TokenCountMeta
  144. *ClaudeConvertInfo
  145. *RerankerInfo
  146. *ResponsesUsageInfo
  147. *ChannelMeta
  148. *TaskRelayInfo
  149. }
  150. func (info *RelayInfo) InitChannelMeta(c *gin.Context) {
  151. channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  152. paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  153. headerOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelHeaderOverride)
  154. apiType, _ := common.ChannelType2APIType(channelType)
  155. channelMeta := &ChannelMeta{
  156. ChannelType: channelType,
  157. ChannelId: common.GetContextKeyInt(c, constant.ContextKeyChannelId),
  158. ChannelIsMultiKey: common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey),
  159. ChannelMultiKeyIndex: common.GetContextKeyInt(c, constant.ContextKeyChannelMultiKeyIndex),
  160. ChannelBaseUrl: common.GetContextKeyString(c, constant.ContextKeyChannelBaseUrl),
  161. ApiType: apiType,
  162. ApiVersion: c.GetString("api_version"),
  163. ApiKey: common.GetContextKeyString(c, constant.ContextKeyChannelKey),
  164. Organization: c.GetString("channel_organization"),
  165. ChannelCreateTime: c.GetInt64("channel_create_time"),
  166. ParamOverride: paramOverride,
  167. HeadersOverride: headerOverride,
  168. UpstreamModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  169. IsModelMapped: false,
  170. SupportStreamOptions: false,
  171. }
  172. if channelType == constant.ChannelTypeAzure {
  173. channelMeta.ApiVersion = GetAPIVersion(c)
  174. }
  175. if channelType == constant.ChannelTypeVertexAi {
  176. channelMeta.ApiVersion = c.GetString("region")
  177. }
  178. channelSetting, ok := common.GetContextKeyType[dto.ChannelSettings](c, constant.ContextKeyChannelSetting)
  179. if ok {
  180. channelMeta.ChannelSetting = channelSetting
  181. }
  182. channelOtherSettings, ok := common.GetContextKeyType[dto.ChannelOtherSettings](c, constant.ContextKeyChannelOtherSetting)
  183. if ok {
  184. channelMeta.ChannelOtherSettings = channelOtherSettings
  185. }
  186. if streamSupportedChannels[channelMeta.ChannelType] {
  187. channelMeta.SupportStreamOptions = true
  188. }
  189. info.ChannelMeta = channelMeta
  190. // reset some fields based on channel meta
  191. // 重置某些字段,例如模型名称等
  192. if info.Request != nil {
  193. info.Request.SetModelName(info.OriginModelName)
  194. }
  195. }
  196. func (info *RelayInfo) ToString() string {
  197. if info == nil {
  198. return "RelayInfo<nil>"
  199. }
  200. // Basic info
  201. b := &strings.Builder{}
  202. fmt.Fprintf(b, "RelayInfo{ ")
  203. fmt.Fprintf(b, "RelayFormat: %s, ", info.RelayFormat)
  204. fmt.Fprintf(b, "RelayMode: %d, ", info.RelayMode)
  205. fmt.Fprintf(b, "IsStream: %t, ", info.IsStream)
  206. fmt.Fprintf(b, "IsPlayground: %t, ", info.IsPlayground)
  207. fmt.Fprintf(b, "RequestURLPath: %q, ", info.RequestURLPath)
  208. fmt.Fprintf(b, "OriginModelName: %q, ", info.OriginModelName)
  209. fmt.Fprintf(b, "EstimatePromptTokens: %d, ", info.estimatePromptTokens)
  210. fmt.Fprintf(b, "ShouldIncludeUsage: %t, ", info.ShouldIncludeUsage)
  211. fmt.Fprintf(b, "DisablePing: %t, ", info.DisablePing)
  212. fmt.Fprintf(b, "SendResponseCount: %d, ", info.SendResponseCount)
  213. fmt.Fprintf(b, "FinalPreConsumedQuota: %d, ", info.FinalPreConsumedQuota)
  214. // User & token info (mask secrets)
  215. fmt.Fprintf(b, "User{ Id: %d, Email: %q, Group: %q, UsingGroup: %q, Quota: %d }, ",
  216. info.UserId, common.MaskEmail(info.UserEmail), info.UserGroup, info.UsingGroup, info.UserQuota)
  217. fmt.Fprintf(b, "Token{ Id: %d, Unlimited: %t, Key: ***masked*** }, ", info.TokenId, info.TokenUnlimited)
  218. // Time info
  219. latencyMs := info.FirstResponseTime.Sub(info.StartTime).Milliseconds()
  220. fmt.Fprintf(b, "Timing{ Start: %s, FirstResponse: %s, LatencyMs: %d }, ",
  221. info.StartTime.Format(time.RFC3339Nano), info.FirstResponseTime.Format(time.RFC3339Nano), latencyMs)
  222. // Audio / realtime
  223. if info.InputAudioFormat != "" || info.OutputAudioFormat != "" || len(info.RealtimeTools) > 0 || info.AudioUsage {
  224. fmt.Fprintf(b, "Realtime{ AudioUsage: %t, InFmt: %q, OutFmt: %q, Tools: %d }, ",
  225. info.AudioUsage, info.InputAudioFormat, info.OutputAudioFormat, len(info.RealtimeTools))
  226. }
  227. // Reasoning
  228. if info.ReasoningEffort != "" {
  229. fmt.Fprintf(b, "ReasoningEffort: %q, ", info.ReasoningEffort)
  230. }
  231. // Price data (non-sensitive)
  232. if info.PriceData.UsePrice {
  233. fmt.Fprintf(b, "PriceData{ %s }, ", info.PriceData.ToSetting())
  234. }
  235. // Channel metadata (mask ApiKey)
  236. if info.ChannelMeta != nil {
  237. cm := info.ChannelMeta
  238. 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*** }, ",
  239. cm.ChannelType, cm.ChannelId, cm.ChannelIsMultiKey, cm.ChannelMultiKeyIndex, cm.ChannelBaseUrl, cm.ApiType, cm.ApiVersion, cm.Organization, cm.ChannelCreateTime, cm.UpstreamModelName, cm.IsModelMapped, cm.SupportStreamOptions)
  240. }
  241. // Responses usage info (non-sensitive)
  242. if info.ResponsesUsageInfo != nil && len(info.ResponsesUsageInfo.BuiltInTools) > 0 {
  243. fmt.Fprintf(b, "ResponsesTools{ ")
  244. first := true
  245. for name, tool := range info.ResponsesUsageInfo.BuiltInTools {
  246. if !first {
  247. fmt.Fprintf(b, ", ")
  248. }
  249. first = false
  250. if tool != nil {
  251. fmt.Fprintf(b, "%s: calls=%d", name, tool.CallCount)
  252. } else {
  253. fmt.Fprintf(b, "%s: calls=0", name)
  254. }
  255. }
  256. fmt.Fprintf(b, " }, ")
  257. }
  258. fmt.Fprintf(b, "}")
  259. return b.String()
  260. }
  261. // 定义支持流式选项的通道类型
  262. var streamSupportedChannels = map[int]bool{
  263. constant.ChannelTypeOpenAI: true,
  264. constant.ChannelTypeAnthropic: true,
  265. constant.ChannelTypeAws: true,
  266. constant.ChannelTypeGemini: true,
  267. constant.ChannelCloudflare: true,
  268. constant.ChannelTypeAzure: true,
  269. constant.ChannelTypeVolcEngine: true,
  270. constant.ChannelTypeOllama: true,
  271. constant.ChannelTypeXai: true,
  272. constant.ChannelTypeDeepSeek: true,
  273. constant.ChannelTypeBaiduV2: true,
  274. constant.ChannelTypeZhipu_v4: true,
  275. constant.ChannelTypeAli: true,
  276. constant.ChannelTypeSubmodel: true,
  277. constant.ChannelTypeCodex: true,
  278. constant.ChannelTypeMoonshot: true,
  279. constant.ChannelTypeMiniMax: true,
  280. constant.ChannelTypeSiliconFlow: true,
  281. }
  282. func GenRelayInfoWs(c *gin.Context, ws *websocket.Conn) *RelayInfo {
  283. info := genBaseRelayInfo(c, nil)
  284. info.RelayFormat = types.RelayFormatOpenAIRealtime
  285. info.ClientWs = ws
  286. info.InputAudioFormat = "pcm16"
  287. info.OutputAudioFormat = "pcm16"
  288. info.IsFirstRequest = true
  289. return info
  290. }
  291. func GenRelayInfoClaude(c *gin.Context, request dto.Request) *RelayInfo {
  292. info := genBaseRelayInfo(c, request)
  293. info.RelayFormat = types.RelayFormatClaude
  294. info.ShouldIncludeUsage = false
  295. info.ClaudeConvertInfo = &ClaudeConvertInfo{
  296. LastMessagesType: LastMessageTypeNone,
  297. }
  298. info.IsClaudeBetaQuery = c.Query("beta") == "true" || isClaudeBetaForced(c)
  299. return info
  300. }
  301. func isClaudeBetaForced(c *gin.Context) bool {
  302. channelOtherSettings, ok := common.GetContextKeyType[dto.ChannelOtherSettings](c, constant.ContextKeyChannelOtherSetting)
  303. return ok && channelOtherSettings.ClaudeBetaQuery
  304. }
  305. func GenRelayInfoRerank(c *gin.Context, request *dto.RerankRequest) *RelayInfo {
  306. info := genBaseRelayInfo(c, request)
  307. info.RelayMode = relayconstant.RelayModeRerank
  308. info.RelayFormat = types.RelayFormatRerank
  309. info.RerankerInfo = &RerankerInfo{
  310. Documents: request.Documents,
  311. ReturnDocuments: request.GetReturnDocuments(),
  312. }
  313. return info
  314. }
  315. func GenRelayInfoOpenAIAudio(c *gin.Context, request dto.Request) *RelayInfo {
  316. info := genBaseRelayInfo(c, request)
  317. info.RelayFormat = types.RelayFormatOpenAIAudio
  318. return info
  319. }
  320. func GenRelayInfoEmbedding(c *gin.Context, request dto.Request) *RelayInfo {
  321. info := genBaseRelayInfo(c, request)
  322. info.RelayFormat = types.RelayFormatEmbedding
  323. return info
  324. }
  325. func GenRelayInfoResponses(c *gin.Context, request *dto.OpenAIResponsesRequest) *RelayInfo {
  326. info := genBaseRelayInfo(c, request)
  327. info.RelayMode = relayconstant.RelayModeResponses
  328. info.RelayFormat = types.RelayFormatOpenAIResponses
  329. info.ResponsesUsageInfo = &ResponsesUsageInfo{
  330. BuiltInTools: make(map[string]*BuildInToolInfo),
  331. }
  332. if len(request.Tools) > 0 {
  333. for _, tool := range request.GetToolsMap() {
  334. toolType := common.Interface2String(tool["type"])
  335. info.ResponsesUsageInfo.BuiltInTools[toolType] = &BuildInToolInfo{
  336. ToolName: toolType,
  337. CallCount: 0,
  338. }
  339. switch toolType {
  340. case dto.BuildInToolWebSearchPreview:
  341. searchContextSize := common.Interface2String(tool["search_context_size"])
  342. if searchContextSize == "" {
  343. searchContextSize = "medium"
  344. }
  345. info.ResponsesUsageInfo.BuiltInTools[toolType].SearchContextSize = searchContextSize
  346. }
  347. }
  348. }
  349. return info
  350. }
  351. func GenRelayInfoGemini(c *gin.Context, request dto.Request) *RelayInfo {
  352. info := genBaseRelayInfo(c, request)
  353. info.RelayFormat = types.RelayFormatGemini
  354. info.ShouldIncludeUsage = false
  355. return info
  356. }
  357. func GenRelayInfoImage(c *gin.Context, request dto.Request) *RelayInfo {
  358. info := genBaseRelayInfo(c, request)
  359. info.RelayFormat = types.RelayFormatOpenAIImage
  360. return info
  361. }
  362. func GenRelayInfoOpenAI(c *gin.Context, request dto.Request) *RelayInfo {
  363. info := genBaseRelayInfo(c, request)
  364. info.RelayFormat = types.RelayFormatOpenAI
  365. return info
  366. }
  367. func genBaseRelayInfo(c *gin.Context, request dto.Request) *RelayInfo {
  368. //channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  369. //channelId := common.GetContextKeyInt(c, constant.ContextKeyChannelId)
  370. //paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  371. tokenGroup := common.GetContextKeyString(c, constant.ContextKeyTokenGroup)
  372. // 当令牌分组为空时,表示使用用户分组
  373. if tokenGroup == "" {
  374. tokenGroup = common.GetContextKeyString(c, constant.ContextKeyUserGroup)
  375. }
  376. startTime := common.GetContextKeyTime(c, constant.ContextKeyRequestStartTime)
  377. if startTime.IsZero() {
  378. startTime = time.Now()
  379. }
  380. isStream := false
  381. if request != nil {
  382. isStream = request.IsStream(c)
  383. }
  384. // firstResponseTime = time.Now() - 1 second
  385. reqId := common.GetContextKeyString(c, common.RequestIdKey)
  386. if reqId == "" {
  387. reqId = common.GetTimeString() + common.GetRandomString(8)
  388. }
  389. info := &RelayInfo{
  390. Request: request,
  391. RequestId: reqId,
  392. UserId: common.GetContextKeyInt(c, constant.ContextKeyUserId),
  393. UsingGroup: common.GetContextKeyString(c, constant.ContextKeyUsingGroup),
  394. UserGroup: common.GetContextKeyString(c, constant.ContextKeyUserGroup),
  395. UserQuota: common.GetContextKeyInt(c, constant.ContextKeyUserQuota),
  396. UserEmail: common.GetContextKeyString(c, constant.ContextKeyUserEmail),
  397. OriginModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  398. TokenId: common.GetContextKeyInt(c, constant.ContextKeyTokenId),
  399. TokenKey: common.GetContextKeyString(c, constant.ContextKeyTokenKey),
  400. TokenUnlimited: common.GetContextKeyBool(c, constant.ContextKeyTokenUnlimited),
  401. TokenGroup: tokenGroup,
  402. isFirstResponse: true,
  403. RelayMode: relayconstant.Path2RelayMode(c.Request.URL.Path),
  404. RequestURLPath: c.Request.URL.String(),
  405. IsStream: isStream,
  406. StartTime: startTime,
  407. FirstResponseTime: startTime.Add(-time.Second),
  408. ThinkingContentInfo: ThinkingContentInfo{
  409. IsFirstThinkingContent: true,
  410. SendLastThinkingContent: false,
  411. },
  412. TokenCountMeta: TokenCountMeta{
  413. //promptTokens: common.GetContextKeyInt(c, constant.ContextKeyPromptTokens),
  414. estimatePromptTokens: common.GetContextKeyInt(c, constant.ContextKeyEstimatedTokens),
  415. },
  416. }
  417. if info.RelayMode == relayconstant.RelayModeUnknown {
  418. info.RelayMode = c.GetInt("relay_mode")
  419. }
  420. if strings.HasPrefix(c.Request.URL.Path, "/pg") {
  421. info.IsPlayground = true
  422. info.RequestURLPath = strings.TrimPrefix(info.RequestURLPath, "/pg")
  423. info.RequestURLPath = "/v1" + info.RequestURLPath
  424. }
  425. userSetting, ok := common.GetContextKeyType[dto.UserSetting](c, constant.ContextKeyUserSetting)
  426. if ok {
  427. info.UserSetting = userSetting
  428. }
  429. return info
  430. }
  431. func GenRelayInfo(c *gin.Context, relayFormat types.RelayFormat, request dto.Request, ws *websocket.Conn) (*RelayInfo, error) {
  432. var info *RelayInfo
  433. var err error
  434. switch relayFormat {
  435. case types.RelayFormatOpenAI:
  436. info = GenRelayInfoOpenAI(c, request)
  437. case types.RelayFormatOpenAIAudio:
  438. info = GenRelayInfoOpenAIAudio(c, request)
  439. case types.RelayFormatOpenAIImage:
  440. info = GenRelayInfoImage(c, request)
  441. case types.RelayFormatOpenAIRealtime:
  442. info = GenRelayInfoWs(c, ws)
  443. case types.RelayFormatClaude:
  444. info = GenRelayInfoClaude(c, request)
  445. case types.RelayFormatRerank:
  446. if request, ok := request.(*dto.RerankRequest); ok {
  447. info = GenRelayInfoRerank(c, request)
  448. break
  449. }
  450. err = errors.New("request is not a RerankRequest")
  451. case types.RelayFormatGemini:
  452. info = GenRelayInfoGemini(c, request)
  453. case types.RelayFormatEmbedding:
  454. info = GenRelayInfoEmbedding(c, request)
  455. case types.RelayFormatOpenAIResponses:
  456. if request, ok := request.(*dto.OpenAIResponsesRequest); ok {
  457. info = GenRelayInfoResponses(c, request)
  458. break
  459. }
  460. err = errors.New("request is not a OpenAIResponsesRequest")
  461. case types.RelayFormatOpenAIResponsesCompaction:
  462. if request, ok := request.(*dto.OpenAIResponsesCompactionRequest); ok {
  463. return GenRelayInfoResponsesCompaction(c, request), nil
  464. }
  465. return nil, errors.New("request is not a OpenAIResponsesCompactionRequest")
  466. case types.RelayFormatTask:
  467. info = genBaseRelayInfo(c, nil)
  468. info.TaskRelayInfo = &TaskRelayInfo{}
  469. case types.RelayFormatMjProxy:
  470. info = genBaseRelayInfo(c, nil)
  471. info.TaskRelayInfo = &TaskRelayInfo{}
  472. default:
  473. err = errors.New("invalid relay format")
  474. }
  475. if err != nil {
  476. return nil, err
  477. }
  478. if info == nil {
  479. return nil, errors.New("failed to build relay info")
  480. }
  481. info.InitRequestConversionChain()
  482. return info, nil
  483. }
  484. func (info *RelayInfo) InitRequestConversionChain() {
  485. if info == nil {
  486. return
  487. }
  488. if len(info.RequestConversionChain) > 0 {
  489. return
  490. }
  491. if info.RelayFormat == "" {
  492. return
  493. }
  494. info.RequestConversionChain = []types.RelayFormat{info.RelayFormat}
  495. }
  496. func (info *RelayInfo) AppendRequestConversion(format types.RelayFormat) {
  497. if info == nil {
  498. return
  499. }
  500. if format == "" {
  501. return
  502. }
  503. if len(info.RequestConversionChain) == 0 {
  504. info.RequestConversionChain = []types.RelayFormat{format}
  505. return
  506. }
  507. last := info.RequestConversionChain[len(info.RequestConversionChain)-1]
  508. if last == format {
  509. return
  510. }
  511. info.RequestConversionChain = append(info.RequestConversionChain, format)
  512. }
  513. func (info *RelayInfo) GetFinalRequestRelayFormat() types.RelayFormat {
  514. if info == nil {
  515. return ""
  516. }
  517. if info.FinalRequestRelayFormat != "" {
  518. return info.FinalRequestRelayFormat
  519. }
  520. if n := len(info.RequestConversionChain); n > 0 {
  521. return info.RequestConversionChain[n-1]
  522. }
  523. return info.RelayFormat
  524. }
  525. func GenRelayInfoResponsesCompaction(c *gin.Context, request *dto.OpenAIResponsesCompactionRequest) *RelayInfo {
  526. info := genBaseRelayInfo(c, request)
  527. if info.RelayMode == relayconstant.RelayModeUnknown {
  528. info.RelayMode = relayconstant.RelayModeResponsesCompact
  529. }
  530. info.RelayFormat = types.RelayFormatOpenAIResponsesCompaction
  531. return info
  532. }
  533. //func (info *RelayInfo) SetPromptTokens(promptTokens int) {
  534. // info.promptTokens = promptTokens
  535. //}
  536. func (info *RelayInfo) SetEstimatePromptTokens(promptTokens int) {
  537. info.estimatePromptTokens = promptTokens
  538. }
  539. func (info *RelayInfo) GetEstimatePromptTokens() int {
  540. return info.estimatePromptTokens
  541. }
  542. func (info *RelayInfo) SetFirstResponseTime() {
  543. if info.isFirstResponse {
  544. info.FirstResponseTime = time.Now()
  545. info.isFirstResponse = false
  546. }
  547. }
  548. func (info *RelayInfo) HasSendResponse() bool {
  549. return info.FirstResponseTime.After(info.StartTime)
  550. }
  551. type TaskRelayInfo struct {
  552. Action string
  553. OriginTaskID string
  554. // PublicTaskID 是提交时预生成的 task_xxxx 格式公开 ID,
  555. // 供 DoResponse 在返回给客户端时使用(避免暴露上游真实 ID)。
  556. PublicTaskID string
  557. ConsumeQuota bool
  558. // LockedChannel holds the full channel object when the request is bound to
  559. // a specific channel (e.g., remix on origin task's channel). Stored as any
  560. // to avoid an import cycle with model; callers type-assert to *model.Channel.
  561. LockedChannel any
  562. }
  563. type TaskSubmitReq struct {
  564. Prompt string `json:"prompt"`
  565. Model string `json:"model,omitempty"`
  566. Mode string `json:"mode,omitempty"`
  567. Image string `json:"image,omitempty"`
  568. Images []string `json:"images,omitempty"`
  569. Size string `json:"size,omitempty"`
  570. Duration int `json:"duration,omitempty"`
  571. Seconds string `json:"seconds,omitempty"`
  572. InputReference string `json:"input_reference,omitempty"`
  573. Metadata map[string]interface{} `json:"metadata,omitempty"`
  574. }
  575. func (t *TaskSubmitReq) GetPrompt() string {
  576. return t.Prompt
  577. }
  578. func (t *TaskSubmitReq) HasImage() bool {
  579. return len(t.Images) > 0
  580. }
  581. func (t *TaskSubmitReq) UnmarshalJSON(data []byte) error {
  582. type Alias TaskSubmitReq
  583. aux := &struct {
  584. Metadata json.RawMessage `json:"metadata,omitempty"`
  585. *Alias
  586. }{
  587. Alias: (*Alias)(t),
  588. }
  589. if err := common.Unmarshal(data, &aux); err != nil {
  590. return err
  591. }
  592. if len(aux.Metadata) > 0 {
  593. var metadataStr string
  594. if err := common.Unmarshal(aux.Metadata, &metadataStr); err == nil && metadataStr != "" {
  595. var metadataObj map[string]interface{}
  596. if err := common.Unmarshal([]byte(metadataStr), &metadataObj); err == nil {
  597. t.Metadata = metadataObj
  598. return nil
  599. }
  600. }
  601. var metadataObj map[string]interface{}
  602. if err := common.Unmarshal(aux.Metadata, &metadataObj); err == nil {
  603. t.Metadata = metadataObj
  604. }
  605. }
  606. return nil
  607. }
  608. func (t *TaskSubmitReq) UnmarshalMetadata(v any) error {
  609. metadata := t.Metadata
  610. if metadata != nil {
  611. metadataBytes, err := common.Marshal(metadata)
  612. if err != nil {
  613. return fmt.Errorf("marshal metadata failed: %w", err)
  614. }
  615. err = common.Unmarshal(metadataBytes, v)
  616. if err != nil {
  617. return fmt.Errorf("unmarshal metadata to target failed: %w", err)
  618. }
  619. }
  620. return nil
  621. }
  622. type TaskInfo struct {
  623. Code int `json:"code"`
  624. TaskID string `json:"task_id"`
  625. Status string `json:"status"`
  626. Reason string `json:"reason,omitempty"`
  627. Url string `json:"url,omitempty"`
  628. RemoteUrl string `json:"remote_url,omitempty"`
  629. Progress string `json:"progress,omitempty"`
  630. CompletionTokens int `json:"completion_tokens,omitempty"` // 用于按倍率计费
  631. TotalTokens int `json:"total_tokens,omitempty"` // 用于按倍率计费
  632. }
  633. func FailTaskInfo(reason string) *TaskInfo {
  634. return &TaskInfo{
  635. Status: "FAILURE",
  636. Reason: reason,
  637. }
  638. }
  639. // RemoveDisabledFields 从请求 JSON 数据中移除渠道设置中禁用的字段
  640. // service_tier: 服务层级字段,可能导致额外计费(OpenAI、Claude、Responses API 支持)
  641. // 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)
  642. // 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)
  643. func RemoveDisabledFields(jsonData []byte, channelOtherSettings dto.ChannelOtherSettings) ([]byte, error) {
  644. var data map[string]interface{}
  645. if err := common.Unmarshal(jsonData, &data); err != nil {
  646. common.SysError("RemoveDisabledFields Unmarshal error :" + err.Error())
  647. return jsonData, nil
  648. }
  649. // 默认移除 service_tier,除非明确允许(避免额外计费风险)
  650. if !channelOtherSettings.AllowServiceTier {
  651. if _, exists := data["service_tier"]; exists {
  652. delete(data, "service_tier")
  653. }
  654. }
  655. // 默认允许 store 透传,除非明确禁用(禁用可能影响 Codex 使用)
  656. if channelOtherSettings.DisableStore {
  657. if _, exists := data["store"]; exists {
  658. delete(data, "store")
  659. }
  660. }
  661. // 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)
  662. if !channelOtherSettings.AllowSafetyIdentifier {
  663. if _, exists := data["safety_identifier"]; exists {
  664. delete(data, "safety_identifier")
  665. }
  666. }
  667. jsonDataAfter, err := common.Marshal(data)
  668. if err != nil {
  669. common.SysError("RemoveDisabledFields Marshal error :" + err.Error())
  670. return jsonData, nil
  671. }
  672. return jsonDataAfter, nil
  673. }
  674. // RemoveGeminiDisabledFields removes disabled fields from Gemini request JSON data
  675. // Currently supports removing functionResponse.id field which Vertex AI does not support
  676. func RemoveGeminiDisabledFields(jsonData []byte) ([]byte, error) {
  677. if !model_setting.GetGeminiSettings().RemoveFunctionResponseIdEnabled {
  678. return jsonData, nil
  679. }
  680. var data map[string]interface{}
  681. if err := common.Unmarshal(jsonData, &data); err != nil {
  682. common.SysError("RemoveGeminiDisabledFields Unmarshal error: " + err.Error())
  683. return jsonData, nil
  684. }
  685. // Process contents array
  686. // Handle both camelCase (functionResponse) and snake_case (function_response)
  687. if contents, ok := data["contents"].([]interface{}); ok {
  688. for _, content := range contents {
  689. if contentMap, ok := content.(map[string]interface{}); ok {
  690. if parts, ok := contentMap["parts"].([]interface{}); ok {
  691. for _, part := range parts {
  692. if partMap, ok := part.(map[string]interface{}); ok {
  693. // Check functionResponse (camelCase)
  694. if funcResp, ok := partMap["functionResponse"].(map[string]interface{}); ok {
  695. delete(funcResp, "id")
  696. }
  697. // Check function_response (snake_case)
  698. if funcResp, ok := partMap["function_response"].(map[string]interface{}); ok {
  699. delete(funcResp, "id")
  700. }
  701. }
  702. }
  703. }
  704. }
  705. }
  706. }
  707. jsonDataAfter, err := common.Marshal(data)
  708. if err != nil {
  709. common.SysError("RemoveGeminiDisabledFields Marshal error: " + err.Error())
  710. return jsonData, nil
  711. }
  712. return jsonDataAfter, nil
  713. }