relay_info.go 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783
  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. // Billing 是计费会话,封装了预扣费/结算/退款的统一生命周期。
  109. // 免费模型和按次计费(MJ/Task)时为 nil。
  110. Billing BillingSettler
  111. // BillingSource indicates whether this request is billed from wallet quota or subscription.
  112. // "" or "wallet" => wallet; "subscription" => subscription
  113. BillingSource string
  114. // SubscriptionId is the user_subscriptions.id used when BillingSource == "subscription"
  115. SubscriptionId int
  116. // SubscriptionPreConsumed is the amount pre-consumed on subscription item (quota units or 1)
  117. SubscriptionPreConsumed int64
  118. // SubscriptionPostDelta is the post-consume delta applied to amount_used (quota units; can be negative).
  119. SubscriptionPostDelta int64
  120. // SubscriptionPlanId / SubscriptionPlanTitle are used for logging/UI display.
  121. SubscriptionPlanId int
  122. SubscriptionPlanTitle string
  123. // RequestId is used for idempotent pre-consume/refund
  124. RequestId string
  125. // SubscriptionAmountTotal / SubscriptionAmountUsedAfterPreConsume are used to compute remaining in logs.
  126. SubscriptionAmountTotal int64
  127. SubscriptionAmountUsedAfterPreConsume int64
  128. IsClaudeBetaQuery bool // /v1/messages?beta=true
  129. IsChannelTest bool // channel test request
  130. PriceData types.PriceData
  131. Request dto.Request
  132. // RequestConversionChain records request format conversions in order, e.g.
  133. // ["openai", "openai_responses"] or ["openai", "claude"].
  134. RequestConversionChain []types.RelayFormat
  135. // 最终请求到上游的格式 TODO: 当前仅设置了Claude
  136. FinalRequestRelayFormat types.RelayFormat
  137. ThinkingContentInfo
  138. TokenCountMeta
  139. *ClaudeConvertInfo
  140. *RerankerInfo
  141. *ResponsesUsageInfo
  142. *ChannelMeta
  143. *TaskRelayInfo
  144. }
  145. func (info *RelayInfo) InitChannelMeta(c *gin.Context) {
  146. channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  147. paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  148. headerOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelHeaderOverride)
  149. apiType, _ := common.ChannelType2APIType(channelType)
  150. channelMeta := &ChannelMeta{
  151. ChannelType: channelType,
  152. ChannelId: common.GetContextKeyInt(c, constant.ContextKeyChannelId),
  153. ChannelIsMultiKey: common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey),
  154. ChannelMultiKeyIndex: common.GetContextKeyInt(c, constant.ContextKeyChannelMultiKeyIndex),
  155. ChannelBaseUrl: common.GetContextKeyString(c, constant.ContextKeyChannelBaseUrl),
  156. ApiType: apiType,
  157. ApiVersion: c.GetString("api_version"),
  158. ApiKey: common.GetContextKeyString(c, constant.ContextKeyChannelKey),
  159. Organization: c.GetString("channel_organization"),
  160. ChannelCreateTime: c.GetInt64("channel_create_time"),
  161. ParamOverride: paramOverride,
  162. HeadersOverride: headerOverride,
  163. UpstreamModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  164. IsModelMapped: false,
  165. SupportStreamOptions: false,
  166. }
  167. if channelType == constant.ChannelTypeAzure {
  168. channelMeta.ApiVersion = GetAPIVersion(c)
  169. }
  170. if channelType == constant.ChannelTypeVertexAi {
  171. channelMeta.ApiVersion = c.GetString("region")
  172. }
  173. channelSetting, ok := common.GetContextKeyType[dto.ChannelSettings](c, constant.ContextKeyChannelSetting)
  174. if ok {
  175. channelMeta.ChannelSetting = channelSetting
  176. }
  177. channelOtherSettings, ok := common.GetContextKeyType[dto.ChannelOtherSettings](c, constant.ContextKeyChannelOtherSetting)
  178. if ok {
  179. channelMeta.ChannelOtherSettings = channelOtherSettings
  180. }
  181. if streamSupportedChannels[channelMeta.ChannelType] {
  182. channelMeta.SupportStreamOptions = true
  183. }
  184. info.ChannelMeta = channelMeta
  185. // reset some fields based on channel meta
  186. // 重置某些字段,例如模型名称等
  187. if info.Request != nil {
  188. info.Request.SetModelName(info.OriginModelName)
  189. }
  190. }
  191. func (info *RelayInfo) ToString() string {
  192. if info == nil {
  193. return "RelayInfo<nil>"
  194. }
  195. // Basic info
  196. b := &strings.Builder{}
  197. fmt.Fprintf(b, "RelayInfo{ ")
  198. fmt.Fprintf(b, "RelayFormat: %s, ", info.RelayFormat)
  199. fmt.Fprintf(b, "RelayMode: %d, ", info.RelayMode)
  200. fmt.Fprintf(b, "IsStream: %t, ", info.IsStream)
  201. fmt.Fprintf(b, "IsPlayground: %t, ", info.IsPlayground)
  202. fmt.Fprintf(b, "RequestURLPath: %q, ", info.RequestURLPath)
  203. fmt.Fprintf(b, "OriginModelName: %q, ", info.OriginModelName)
  204. fmt.Fprintf(b, "EstimatePromptTokens: %d, ", info.estimatePromptTokens)
  205. fmt.Fprintf(b, "ShouldIncludeUsage: %t, ", info.ShouldIncludeUsage)
  206. fmt.Fprintf(b, "DisablePing: %t, ", info.DisablePing)
  207. fmt.Fprintf(b, "SendResponseCount: %d, ", info.SendResponseCount)
  208. fmt.Fprintf(b, "FinalPreConsumedQuota: %d, ", info.FinalPreConsumedQuota)
  209. // User & token info (mask secrets)
  210. fmt.Fprintf(b, "User{ Id: %d, Email: %q, Group: %q, UsingGroup: %q, Quota: %d }, ",
  211. info.UserId, common.MaskEmail(info.UserEmail), info.UserGroup, info.UsingGroup, info.UserQuota)
  212. fmt.Fprintf(b, "Token{ Id: %d, Unlimited: %t, Key: ***masked*** }, ", info.TokenId, info.TokenUnlimited)
  213. // Time info
  214. latencyMs := info.FirstResponseTime.Sub(info.StartTime).Milliseconds()
  215. fmt.Fprintf(b, "Timing{ Start: %s, FirstResponse: %s, LatencyMs: %d }, ",
  216. info.StartTime.Format(time.RFC3339Nano), info.FirstResponseTime.Format(time.RFC3339Nano), latencyMs)
  217. // Audio / realtime
  218. if info.InputAudioFormat != "" || info.OutputAudioFormat != "" || len(info.RealtimeTools) > 0 || info.AudioUsage {
  219. fmt.Fprintf(b, "Realtime{ AudioUsage: %t, InFmt: %q, OutFmt: %q, Tools: %d }, ",
  220. info.AudioUsage, info.InputAudioFormat, info.OutputAudioFormat, len(info.RealtimeTools))
  221. }
  222. // Reasoning
  223. if info.ReasoningEffort != "" {
  224. fmt.Fprintf(b, "ReasoningEffort: %q, ", info.ReasoningEffort)
  225. }
  226. // Price data (non-sensitive)
  227. if info.PriceData.UsePrice {
  228. fmt.Fprintf(b, "PriceData{ %s }, ", info.PriceData.ToSetting())
  229. }
  230. // Channel metadata (mask ApiKey)
  231. if info.ChannelMeta != nil {
  232. cm := info.ChannelMeta
  233. 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*** }, ",
  234. cm.ChannelType, cm.ChannelId, cm.ChannelIsMultiKey, cm.ChannelMultiKeyIndex, cm.ChannelBaseUrl, cm.ApiType, cm.ApiVersion, cm.Organization, cm.ChannelCreateTime, cm.UpstreamModelName, cm.IsModelMapped, cm.SupportStreamOptions)
  235. }
  236. // Responses usage info (non-sensitive)
  237. if info.ResponsesUsageInfo != nil && len(info.ResponsesUsageInfo.BuiltInTools) > 0 {
  238. fmt.Fprintf(b, "ResponsesTools{ ")
  239. first := true
  240. for name, tool := range info.ResponsesUsageInfo.BuiltInTools {
  241. if !first {
  242. fmt.Fprintf(b, ", ")
  243. }
  244. first = false
  245. if tool != nil {
  246. fmt.Fprintf(b, "%s: calls=%d", name, tool.CallCount)
  247. } else {
  248. fmt.Fprintf(b, "%s: calls=0", name)
  249. }
  250. }
  251. fmt.Fprintf(b, " }, ")
  252. }
  253. fmt.Fprintf(b, "}")
  254. return b.String()
  255. }
  256. // 定义支持流式选项的通道类型
  257. var streamSupportedChannels = map[int]bool{
  258. constant.ChannelTypeOpenAI: true,
  259. constant.ChannelTypeAnthropic: true,
  260. constant.ChannelTypeAws: true,
  261. constant.ChannelTypeGemini: true,
  262. constant.ChannelCloudflare: true,
  263. constant.ChannelTypeAzure: true,
  264. constant.ChannelTypeVolcEngine: true,
  265. constant.ChannelTypeOllama: true,
  266. constant.ChannelTypeXai: true,
  267. constant.ChannelTypeDeepSeek: true,
  268. constant.ChannelTypeBaiduV2: true,
  269. constant.ChannelTypeZhipu_v4: true,
  270. constant.ChannelTypeAli: true,
  271. constant.ChannelTypeSubmodel: true,
  272. constant.ChannelTypeCodex: true,
  273. constant.ChannelTypeMoonshot: true,
  274. constant.ChannelTypeMiniMax: true,
  275. constant.ChannelTypeSiliconFlow: true,
  276. }
  277. func GenRelayInfoWs(c *gin.Context, ws *websocket.Conn) *RelayInfo {
  278. info := genBaseRelayInfo(c, nil)
  279. info.RelayFormat = types.RelayFormatOpenAIRealtime
  280. info.ClientWs = ws
  281. info.InputAudioFormat = "pcm16"
  282. info.OutputAudioFormat = "pcm16"
  283. info.IsFirstRequest = true
  284. return info
  285. }
  286. func GenRelayInfoClaude(c *gin.Context, request dto.Request) *RelayInfo {
  287. info := genBaseRelayInfo(c, request)
  288. info.RelayFormat = types.RelayFormatClaude
  289. info.ShouldIncludeUsage = false
  290. info.ClaudeConvertInfo = &ClaudeConvertInfo{
  291. LastMessagesType: LastMessageTypeNone,
  292. }
  293. info.IsClaudeBetaQuery = c.Query("beta") == "true" || isClaudeBetaForced(c)
  294. return info
  295. }
  296. func isClaudeBetaForced(c *gin.Context) bool {
  297. channelOtherSettings, ok := common.GetContextKeyType[dto.ChannelOtherSettings](c, constant.ContextKeyChannelOtherSetting)
  298. return ok && channelOtherSettings.ClaudeBetaQuery
  299. }
  300. func GenRelayInfoRerank(c *gin.Context, request *dto.RerankRequest) *RelayInfo {
  301. info := genBaseRelayInfo(c, request)
  302. info.RelayMode = relayconstant.RelayModeRerank
  303. info.RelayFormat = types.RelayFormatRerank
  304. info.RerankerInfo = &RerankerInfo{
  305. Documents: request.Documents,
  306. ReturnDocuments: request.GetReturnDocuments(),
  307. }
  308. return info
  309. }
  310. func GenRelayInfoOpenAIAudio(c *gin.Context, request dto.Request) *RelayInfo {
  311. info := genBaseRelayInfo(c, request)
  312. info.RelayFormat = types.RelayFormatOpenAIAudio
  313. return info
  314. }
  315. func GenRelayInfoEmbedding(c *gin.Context, request dto.Request) *RelayInfo {
  316. info := genBaseRelayInfo(c, request)
  317. info.RelayFormat = types.RelayFormatEmbedding
  318. return info
  319. }
  320. func GenRelayInfoResponses(c *gin.Context, request *dto.OpenAIResponsesRequest) *RelayInfo {
  321. info := genBaseRelayInfo(c, request)
  322. info.RelayMode = relayconstant.RelayModeResponses
  323. info.RelayFormat = types.RelayFormatOpenAIResponses
  324. info.ResponsesUsageInfo = &ResponsesUsageInfo{
  325. BuiltInTools: make(map[string]*BuildInToolInfo),
  326. }
  327. if len(request.Tools) > 0 {
  328. for _, tool := range request.GetToolsMap() {
  329. toolType := common.Interface2String(tool["type"])
  330. info.ResponsesUsageInfo.BuiltInTools[toolType] = &BuildInToolInfo{
  331. ToolName: toolType,
  332. CallCount: 0,
  333. }
  334. switch toolType {
  335. case dto.BuildInToolWebSearchPreview:
  336. searchContextSize := common.Interface2String(tool["search_context_size"])
  337. if searchContextSize == "" {
  338. searchContextSize = "medium"
  339. }
  340. info.ResponsesUsageInfo.BuiltInTools[toolType].SearchContextSize = searchContextSize
  341. }
  342. }
  343. }
  344. return info
  345. }
  346. func GenRelayInfoGemini(c *gin.Context, request dto.Request) *RelayInfo {
  347. info := genBaseRelayInfo(c, request)
  348. info.RelayFormat = types.RelayFormatGemini
  349. info.ShouldIncludeUsage = false
  350. return info
  351. }
  352. func GenRelayInfoImage(c *gin.Context, request dto.Request) *RelayInfo {
  353. info := genBaseRelayInfo(c, request)
  354. info.RelayFormat = types.RelayFormatOpenAIImage
  355. return info
  356. }
  357. func GenRelayInfoOpenAI(c *gin.Context, request dto.Request) *RelayInfo {
  358. info := genBaseRelayInfo(c, request)
  359. info.RelayFormat = types.RelayFormatOpenAI
  360. return info
  361. }
  362. func genBaseRelayInfo(c *gin.Context, request dto.Request) *RelayInfo {
  363. //channelType := common.GetContextKeyInt(c, constant.ContextKeyChannelType)
  364. //channelId := common.GetContextKeyInt(c, constant.ContextKeyChannelId)
  365. //paramOverride := common.GetContextKeyStringMap(c, constant.ContextKeyChannelParamOverride)
  366. tokenGroup := common.GetContextKeyString(c, constant.ContextKeyTokenGroup)
  367. // 当令牌分组为空时,表示使用用户分组
  368. if tokenGroup == "" {
  369. tokenGroup = common.GetContextKeyString(c, constant.ContextKeyUserGroup)
  370. }
  371. startTime := common.GetContextKeyTime(c, constant.ContextKeyRequestStartTime)
  372. if startTime.IsZero() {
  373. startTime = time.Now()
  374. }
  375. isStream := false
  376. if request != nil {
  377. isStream = request.IsStream(c)
  378. }
  379. // firstResponseTime = time.Now() - 1 second
  380. reqId := common.GetContextKeyString(c, common.RequestIdKey)
  381. if reqId == "" {
  382. reqId = common.GetTimeString() + common.GetRandomString(8)
  383. }
  384. info := &RelayInfo{
  385. Request: request,
  386. RequestId: reqId,
  387. UserId: common.GetContextKeyInt(c, constant.ContextKeyUserId),
  388. UsingGroup: common.GetContextKeyString(c, constant.ContextKeyUsingGroup),
  389. UserGroup: common.GetContextKeyString(c, constant.ContextKeyUserGroup),
  390. UserQuota: common.GetContextKeyInt(c, constant.ContextKeyUserQuota),
  391. UserEmail: common.GetContextKeyString(c, constant.ContextKeyUserEmail),
  392. OriginModelName: common.GetContextKeyString(c, constant.ContextKeyOriginalModel),
  393. TokenId: common.GetContextKeyInt(c, constant.ContextKeyTokenId),
  394. TokenKey: common.GetContextKeyString(c, constant.ContextKeyTokenKey),
  395. TokenUnlimited: common.GetContextKeyBool(c, constant.ContextKeyTokenUnlimited),
  396. TokenGroup: tokenGroup,
  397. isFirstResponse: true,
  398. RelayMode: relayconstant.Path2RelayMode(c.Request.URL.Path),
  399. RequestURLPath: c.Request.URL.String(),
  400. IsStream: isStream,
  401. StartTime: startTime,
  402. FirstResponseTime: startTime.Add(-time.Second),
  403. ThinkingContentInfo: ThinkingContentInfo{
  404. IsFirstThinkingContent: true,
  405. SendLastThinkingContent: false,
  406. },
  407. TokenCountMeta: TokenCountMeta{
  408. //promptTokens: common.GetContextKeyInt(c, constant.ContextKeyPromptTokens),
  409. estimatePromptTokens: common.GetContextKeyInt(c, constant.ContextKeyEstimatedTokens),
  410. },
  411. }
  412. if info.RelayMode == relayconstant.RelayModeUnknown {
  413. info.RelayMode = c.GetInt("relay_mode")
  414. }
  415. if strings.HasPrefix(c.Request.URL.Path, "/pg") {
  416. info.IsPlayground = true
  417. info.RequestURLPath = strings.TrimPrefix(info.RequestURLPath, "/pg")
  418. info.RequestURLPath = "/v1" + info.RequestURLPath
  419. }
  420. userSetting, ok := common.GetContextKeyType[dto.UserSetting](c, constant.ContextKeyUserSetting)
  421. if ok {
  422. info.UserSetting = userSetting
  423. }
  424. return info
  425. }
  426. func GenRelayInfo(c *gin.Context, relayFormat types.RelayFormat, request dto.Request, ws *websocket.Conn) (*RelayInfo, error) {
  427. var info *RelayInfo
  428. var err error
  429. switch relayFormat {
  430. case types.RelayFormatOpenAI:
  431. info = GenRelayInfoOpenAI(c, request)
  432. case types.RelayFormatOpenAIAudio:
  433. info = GenRelayInfoOpenAIAudio(c, request)
  434. case types.RelayFormatOpenAIImage:
  435. info = GenRelayInfoImage(c, request)
  436. case types.RelayFormatOpenAIRealtime:
  437. info = GenRelayInfoWs(c, ws)
  438. case types.RelayFormatClaude:
  439. info = GenRelayInfoClaude(c, request)
  440. case types.RelayFormatRerank:
  441. if request, ok := request.(*dto.RerankRequest); ok {
  442. info = GenRelayInfoRerank(c, request)
  443. break
  444. }
  445. err = errors.New("request is not a RerankRequest")
  446. case types.RelayFormatGemini:
  447. info = GenRelayInfoGemini(c, request)
  448. case types.RelayFormatEmbedding:
  449. info = GenRelayInfoEmbedding(c, request)
  450. case types.RelayFormatOpenAIResponses:
  451. if request, ok := request.(*dto.OpenAIResponsesRequest); ok {
  452. info = GenRelayInfoResponses(c, request)
  453. break
  454. }
  455. err = errors.New("request is not a OpenAIResponsesRequest")
  456. case types.RelayFormatOpenAIResponsesCompaction:
  457. if request, ok := request.(*dto.OpenAIResponsesCompactionRequest); ok {
  458. return GenRelayInfoResponsesCompaction(c, request), nil
  459. }
  460. return nil, errors.New("request is not a OpenAIResponsesCompactionRequest")
  461. case types.RelayFormatTask:
  462. info = genBaseRelayInfo(c, nil)
  463. case types.RelayFormatMjProxy:
  464. info = genBaseRelayInfo(c, nil)
  465. default:
  466. err = errors.New("invalid relay format")
  467. }
  468. if err != nil {
  469. return nil, err
  470. }
  471. if info == nil {
  472. return nil, errors.New("failed to build relay info")
  473. }
  474. info.InitRequestConversionChain()
  475. return info, nil
  476. }
  477. func (info *RelayInfo) InitRequestConversionChain() {
  478. if info == nil {
  479. return
  480. }
  481. if len(info.RequestConversionChain) > 0 {
  482. return
  483. }
  484. if info.RelayFormat == "" {
  485. return
  486. }
  487. info.RequestConversionChain = []types.RelayFormat{info.RelayFormat}
  488. }
  489. func (info *RelayInfo) AppendRequestConversion(format types.RelayFormat) {
  490. if info == nil {
  491. return
  492. }
  493. if format == "" {
  494. return
  495. }
  496. if len(info.RequestConversionChain) == 0 {
  497. info.RequestConversionChain = []types.RelayFormat{format}
  498. return
  499. }
  500. last := info.RequestConversionChain[len(info.RequestConversionChain)-1]
  501. if last == format {
  502. return
  503. }
  504. info.RequestConversionChain = append(info.RequestConversionChain, format)
  505. }
  506. func GenRelayInfoResponsesCompaction(c *gin.Context, request *dto.OpenAIResponsesCompactionRequest) *RelayInfo {
  507. info := genBaseRelayInfo(c, request)
  508. if info.RelayMode == relayconstant.RelayModeUnknown {
  509. info.RelayMode = relayconstant.RelayModeResponsesCompact
  510. }
  511. info.RelayFormat = types.RelayFormatOpenAIResponsesCompaction
  512. return info
  513. }
  514. //func (info *RelayInfo) SetPromptTokens(promptTokens int) {
  515. // info.promptTokens = promptTokens
  516. //}
  517. func (info *RelayInfo) SetEstimatePromptTokens(promptTokens int) {
  518. info.estimatePromptTokens = promptTokens
  519. }
  520. func (info *RelayInfo) GetEstimatePromptTokens() int {
  521. return info.estimatePromptTokens
  522. }
  523. func (info *RelayInfo) SetFirstResponseTime() {
  524. if info.isFirstResponse {
  525. info.FirstResponseTime = time.Now()
  526. info.isFirstResponse = false
  527. }
  528. }
  529. func (info *RelayInfo) HasSendResponse() bool {
  530. return info.FirstResponseTime.After(info.StartTime)
  531. }
  532. type TaskRelayInfo struct {
  533. Action string
  534. OriginTaskID string
  535. ConsumeQuota bool
  536. }
  537. type TaskSubmitReq struct {
  538. Prompt string `json:"prompt"`
  539. Model string `json:"model,omitempty"`
  540. Mode string `json:"mode,omitempty"`
  541. Image string `json:"image,omitempty"`
  542. Images []string `json:"images,omitempty"`
  543. Size string `json:"size,omitempty"`
  544. Duration int `json:"duration,omitempty"`
  545. Seconds string `json:"seconds,omitempty"`
  546. InputReference string `json:"input_reference,omitempty"`
  547. Metadata map[string]interface{} `json:"metadata,omitempty"`
  548. }
  549. func (t *TaskSubmitReq) GetPrompt() string {
  550. return t.Prompt
  551. }
  552. func (t *TaskSubmitReq) HasImage() bool {
  553. return len(t.Images) > 0
  554. }
  555. func (t *TaskSubmitReq) UnmarshalJSON(data []byte) error {
  556. type Alias TaskSubmitReq
  557. aux := &struct {
  558. Metadata json.RawMessage `json:"metadata,omitempty"`
  559. *Alias
  560. }{
  561. Alias: (*Alias)(t),
  562. }
  563. if err := common.Unmarshal(data, &aux); err != nil {
  564. return err
  565. }
  566. if len(aux.Metadata) > 0 {
  567. var metadataStr string
  568. if err := common.Unmarshal(aux.Metadata, &metadataStr); err == nil && metadataStr != "" {
  569. var metadataObj map[string]interface{}
  570. if err := common.Unmarshal([]byte(metadataStr), &metadataObj); err == nil {
  571. t.Metadata = metadataObj
  572. return nil
  573. }
  574. }
  575. var metadataObj map[string]interface{}
  576. if err := common.Unmarshal(aux.Metadata, &metadataObj); err == nil {
  577. t.Metadata = metadataObj
  578. }
  579. }
  580. return nil
  581. }
  582. func (t *TaskSubmitReq) UnmarshalMetadata(v any) error {
  583. metadata := t.Metadata
  584. if metadata != nil {
  585. metadataBytes, err := json.Marshal(metadata)
  586. if err != nil {
  587. return fmt.Errorf("marshal metadata failed: %w", err)
  588. }
  589. err = json.Unmarshal(metadataBytes, v)
  590. if err != nil {
  591. return fmt.Errorf("unmarshal metadata to target failed: %w", err)
  592. }
  593. }
  594. return nil
  595. }
  596. type TaskInfo struct {
  597. Code int `json:"code"`
  598. TaskID string `json:"task_id"`
  599. Status string `json:"status"`
  600. Reason string `json:"reason,omitempty"`
  601. Url string `json:"url,omitempty"`
  602. RemoteUrl string `json:"remote_url,omitempty"`
  603. Progress string `json:"progress,omitempty"`
  604. CompletionTokens int `json:"completion_tokens,omitempty"` // 用于按倍率计费
  605. TotalTokens int `json:"total_tokens,omitempty"` // 用于按倍率计费
  606. }
  607. func FailTaskInfo(reason string) *TaskInfo {
  608. return &TaskInfo{
  609. Status: "FAILURE",
  610. Reason: reason,
  611. }
  612. }
  613. // RemoveDisabledFields 从请求 JSON 数据中移除渠道设置中禁用的字段
  614. // service_tier: 服务层级字段,可能导致额外计费(OpenAI、Claude、Responses API 支持)
  615. // 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)
  616. // 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)
  617. func RemoveDisabledFields(jsonData []byte, channelOtherSettings dto.ChannelOtherSettings) ([]byte, error) {
  618. var data map[string]interface{}
  619. if err := common.Unmarshal(jsonData, &data); err != nil {
  620. common.SysError("RemoveDisabledFields Unmarshal error :" + err.Error())
  621. return jsonData, nil
  622. }
  623. // 默认移除 service_tier,除非明确允许(避免额外计费风险)
  624. if !channelOtherSettings.AllowServiceTier {
  625. if _, exists := data["service_tier"]; exists {
  626. delete(data, "service_tier")
  627. }
  628. }
  629. // 默认允许 store 透传,除非明确禁用(禁用可能影响 Codex 使用)
  630. if channelOtherSettings.DisableStore {
  631. if _, exists := data["store"]; exists {
  632. delete(data, "store")
  633. }
  634. }
  635. // 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)
  636. if !channelOtherSettings.AllowSafetyIdentifier {
  637. if _, exists := data["safety_identifier"]; exists {
  638. delete(data, "safety_identifier")
  639. }
  640. }
  641. jsonDataAfter, err := common.Marshal(data)
  642. if err != nil {
  643. common.SysError("RemoveDisabledFields Marshal error :" + err.Error())
  644. return jsonData, nil
  645. }
  646. return jsonDataAfter, nil
  647. }
  648. // RemoveGeminiDisabledFields removes disabled fields from Gemini request JSON data
  649. // Currently supports removing functionResponse.id field which Vertex AI does not support
  650. func RemoveGeminiDisabledFields(jsonData []byte) ([]byte, error) {
  651. if !model_setting.GetGeminiSettings().RemoveFunctionResponseIdEnabled {
  652. return jsonData, nil
  653. }
  654. var data map[string]interface{}
  655. if err := common.Unmarshal(jsonData, &data); err != nil {
  656. common.SysError("RemoveGeminiDisabledFields Unmarshal error: " + err.Error())
  657. return jsonData, nil
  658. }
  659. // Process contents array
  660. // Handle both camelCase (functionResponse) and snake_case (function_response)
  661. if contents, ok := data["contents"].([]interface{}); ok {
  662. for _, content := range contents {
  663. if contentMap, ok := content.(map[string]interface{}); ok {
  664. if parts, ok := contentMap["parts"].([]interface{}); ok {
  665. for _, part := range parts {
  666. if partMap, ok := part.(map[string]interface{}); ok {
  667. // Check functionResponse (camelCase)
  668. if funcResp, ok := partMap["functionResponse"].(map[string]interface{}); ok {
  669. delete(funcResp, "id")
  670. }
  671. // Check function_response (snake_case)
  672. if funcResp, ok := partMap["function_response"].(map[string]interface{}); ok {
  673. delete(funcResp, "id")
  674. }
  675. }
  676. }
  677. }
  678. }
  679. }
  680. }
  681. jsonDataAfter, err := common.Marshal(data)
  682. if err != nil {
  683. common.SysError("RemoveGeminiDisabledFields Marshal error: " + err.Error())
  684. return jsonData, nil
  685. }
  686. return jsonDataAfter, nil
  687. }