relay-claude.go 31 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915
  1. package claude
  2. import (
  3. "encoding/json"
  4. "fmt"
  5. "io"
  6. "net/http"
  7. "strings"
  8. "github.com/QuantumNous/new-api/common"
  9. "github.com/QuantumNous/new-api/constant"
  10. "github.com/QuantumNous/new-api/dto"
  11. "github.com/QuantumNous/new-api/logger"
  12. "github.com/QuantumNous/new-api/relay/channel/openrouter"
  13. relaycommon "github.com/QuantumNous/new-api/relay/common"
  14. "github.com/QuantumNous/new-api/relay/helper"
  15. "github.com/QuantumNous/new-api/relay/reasonmap"
  16. "github.com/QuantumNous/new-api/service"
  17. "github.com/QuantumNous/new-api/setting/model_setting"
  18. "github.com/QuantumNous/new-api/setting/reasoning"
  19. "github.com/QuantumNous/new-api/types"
  20. "github.com/gin-gonic/gin"
  21. "github.com/tidwall/gjson"
  22. "github.com/tidwall/sjson"
  23. )
  24. const (
  25. WebSearchMaxUsesLow = 1
  26. WebSearchMaxUsesMedium = 5
  27. WebSearchMaxUsesHigh = 10
  28. )
  29. func stopReasonClaude2OpenAI(reason string) string {
  30. return reasonmap.ClaudeStopReasonToOpenAIFinishReason(reason)
  31. }
  32. func maybeMarkClaudeRefusal(c *gin.Context, stopReason string) {
  33. if c == nil {
  34. return
  35. }
  36. if strings.EqualFold(stopReason, "refusal") {
  37. common.SetContextKey(c, constant.ContextKeyAdminRejectReason, "claude_stop_reason=refusal")
  38. }
  39. }
  40. func RequestOpenAI2ClaudeMessage(c *gin.Context, textRequest dto.GeneralOpenAIRequest) (*dto.ClaudeRequest, error) {
  41. claudeTools := make([]any, 0, len(textRequest.Tools))
  42. for _, tool := range textRequest.Tools {
  43. if params, ok := tool.Function.Parameters.(map[string]any); ok {
  44. claudeTool := dto.Tool{
  45. Name: tool.Function.Name,
  46. Description: tool.Function.Description,
  47. }
  48. claudeTool.InputSchema = make(map[string]interface{})
  49. if params["type"] != nil {
  50. claudeTool.InputSchema["type"] = params["type"].(string)
  51. }
  52. claudeTool.InputSchema["properties"] = params["properties"]
  53. claudeTool.InputSchema["required"] = params["required"]
  54. for s, a := range params {
  55. if s == "type" || s == "properties" || s == "required" {
  56. continue
  57. }
  58. claudeTool.InputSchema[s] = a
  59. }
  60. claudeTools = append(claudeTools, &claudeTool)
  61. }
  62. }
  63. // Web search tool
  64. // https://docs.anthropic.com/en/docs/agents-and-tools/tool-use/web-search-tool
  65. if textRequest.WebSearchOptions != nil {
  66. webSearchTool := dto.ClaudeWebSearchTool{
  67. Type: "web_search_20250305",
  68. Name: "web_search",
  69. }
  70. // 处理 user_location
  71. if textRequest.WebSearchOptions.UserLocation != nil {
  72. anthropicUserLocation := &dto.ClaudeWebSearchUserLocation{
  73. Type: "approximate", // 固定为 "approximate"
  74. }
  75. // 解析 UserLocation JSON
  76. var userLocationMap map[string]interface{}
  77. if err := json.Unmarshal(textRequest.WebSearchOptions.UserLocation, &userLocationMap); err == nil {
  78. // 检查是否有 approximate 字段
  79. if approximateData, ok := userLocationMap["approximate"].(map[string]interface{}); ok {
  80. if timezone, ok := approximateData["timezone"].(string); ok && timezone != "" {
  81. anthropicUserLocation.Timezone = timezone
  82. }
  83. if country, ok := approximateData["country"].(string); ok && country != "" {
  84. anthropicUserLocation.Country = country
  85. }
  86. if region, ok := approximateData["region"].(string); ok && region != "" {
  87. anthropicUserLocation.Region = region
  88. }
  89. if city, ok := approximateData["city"].(string); ok && city != "" {
  90. anthropicUserLocation.City = city
  91. }
  92. }
  93. }
  94. webSearchTool.UserLocation = anthropicUserLocation
  95. }
  96. // 处理 search_context_size 转换为 max_uses
  97. if textRequest.WebSearchOptions.SearchContextSize != "" {
  98. switch textRequest.WebSearchOptions.SearchContextSize {
  99. case "low":
  100. webSearchTool.MaxUses = WebSearchMaxUsesLow
  101. case "medium":
  102. webSearchTool.MaxUses = WebSearchMaxUsesMedium
  103. case "high":
  104. webSearchTool.MaxUses = WebSearchMaxUsesHigh
  105. }
  106. }
  107. claudeTools = append(claudeTools, &webSearchTool)
  108. }
  109. claudeRequest := dto.ClaudeRequest{
  110. Model: textRequest.Model,
  111. StopSequences: nil,
  112. Temperature: textRequest.Temperature,
  113. Tools: claudeTools,
  114. }
  115. if maxTokens := textRequest.GetMaxTokens(); maxTokens > 0 {
  116. claudeRequest.MaxTokens = common.GetPointer(maxTokens)
  117. }
  118. if textRequest.TopP != nil {
  119. claudeRequest.TopP = common.GetPointer(*textRequest.TopP)
  120. }
  121. if textRequest.TopK != nil {
  122. claudeRequest.TopK = common.GetPointer(*textRequest.TopK)
  123. }
  124. if textRequest.IsStream(nil) {
  125. claudeRequest.Stream = common.GetPointer(true)
  126. }
  127. // 处理 tool_choice 和 parallel_tool_calls
  128. if textRequest.ToolChoice != nil || textRequest.ParallelTooCalls != nil {
  129. claudeToolChoice := mapToolChoice(textRequest.ToolChoice, textRequest.ParallelTooCalls)
  130. if claudeToolChoice != nil {
  131. claudeRequest.ToolChoice = claudeToolChoice
  132. }
  133. }
  134. if claudeRequest.MaxTokens == nil || *claudeRequest.MaxTokens == 0 {
  135. defaultMaxTokens := uint(model_setting.GetClaudeSettings().GetDefaultMaxTokens(textRequest.Model))
  136. claudeRequest.MaxTokens = &defaultMaxTokens
  137. }
  138. if baseModel, effortLevel, ok := reasoning.TrimEffortSuffix(textRequest.Model); ok && effortLevel != "" &&
  139. strings.HasPrefix(textRequest.Model, "claude-opus-4-6") {
  140. claudeRequest.Model = baseModel
  141. claudeRequest.Thinking = &dto.Thinking{
  142. Type: "adaptive",
  143. }
  144. claudeRequest.OutputConfig = json.RawMessage(fmt.Sprintf(`{"effort":"%s"}`, effortLevel))
  145. claudeRequest.TopP = common.GetPointer[float64](0)
  146. claudeRequest.Temperature = common.GetPointer[float64](1.0)
  147. } else if model_setting.GetClaudeSettings().ThinkingAdapterEnabled &&
  148. strings.HasSuffix(textRequest.Model, "-thinking") {
  149. // 因为BudgetTokens 必须大于1024
  150. if claudeRequest.MaxTokens == nil || *claudeRequest.MaxTokens < 1280 {
  151. claudeRequest.MaxTokens = common.GetPointer[uint](1280)
  152. }
  153. // BudgetTokens 为 max_tokens 的 80%
  154. claudeRequest.Thinking = &dto.Thinking{
  155. Type: "enabled",
  156. BudgetTokens: common.GetPointer[int](int(float64(*claudeRequest.MaxTokens) * model_setting.GetClaudeSettings().ThinkingAdapterBudgetTokensPercentage)),
  157. }
  158. // TODO: 临时处理
  159. // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerations-when-using-extended-thinking
  160. claudeRequest.TopP = common.GetPointer[float64](0)
  161. claudeRequest.Temperature = common.GetPointer[float64](1.0)
  162. if !model_setting.ShouldPreserveThinkingSuffix(textRequest.Model) {
  163. claudeRequest.Model = strings.TrimSuffix(textRequest.Model, "-thinking")
  164. }
  165. }
  166. if textRequest.ReasoningEffort != "" {
  167. switch textRequest.ReasoningEffort {
  168. case "low":
  169. claudeRequest.Thinking = &dto.Thinking{
  170. Type: "enabled",
  171. BudgetTokens: common.GetPointer[int](1280),
  172. }
  173. case "medium":
  174. claudeRequest.Thinking = &dto.Thinking{
  175. Type: "enabled",
  176. BudgetTokens: common.GetPointer[int](2048),
  177. }
  178. case "high":
  179. claudeRequest.Thinking = &dto.Thinking{
  180. Type: "enabled",
  181. BudgetTokens: common.GetPointer[int](4096),
  182. }
  183. }
  184. }
  185. // 指定了 reasoning 参数,覆盖 budgetTokens
  186. if textRequest.Reasoning != nil {
  187. var reasoning openrouter.RequestReasoning
  188. if err := common.Unmarshal(textRequest.Reasoning, &reasoning); err != nil {
  189. return nil, err
  190. }
  191. budgetTokens := reasoning.MaxTokens
  192. if budgetTokens > 0 {
  193. claudeRequest.Thinking = &dto.Thinking{
  194. Type: "enabled",
  195. BudgetTokens: &budgetTokens,
  196. }
  197. }
  198. }
  199. if textRequest.Stop != nil {
  200. // stop maybe string/array string, convert to array string
  201. switch textRequest.Stop.(type) {
  202. case string:
  203. claudeRequest.StopSequences = []string{textRequest.Stop.(string)}
  204. case []interface{}:
  205. stopSequences := make([]string, 0)
  206. for _, stop := range textRequest.Stop.([]interface{}) {
  207. stopSequences = append(stopSequences, stop.(string))
  208. }
  209. claudeRequest.StopSequences = stopSequences
  210. }
  211. }
  212. formatMessages := make([]dto.Message, 0)
  213. lastMessage := dto.Message{
  214. Role: "tool",
  215. }
  216. for i, message := range textRequest.Messages {
  217. if message.Role == "" {
  218. textRequest.Messages[i].Role = "user"
  219. }
  220. fmtMessage := dto.Message{
  221. Role: message.Role,
  222. Content: message.Content,
  223. }
  224. if message.Role == "tool" {
  225. fmtMessage.ToolCallId = message.ToolCallId
  226. }
  227. if message.Role == "assistant" && message.ToolCalls != nil {
  228. fmtMessage.ToolCalls = message.ToolCalls
  229. if message.IsStringContent() && message.StringContent() == "" {
  230. fmtMessage.SetNullContent()
  231. }
  232. }
  233. if lastMessage.Role == message.Role && lastMessage.Role != "tool" {
  234. if lastMessage.IsStringContent() && message.IsStringContent() {
  235. fmtMessage.SetStringContent(strings.Trim(fmt.Sprintf("%s %s", lastMessage.StringContent(), message.StringContent()), "\""))
  236. // delete last message
  237. formatMessages = formatMessages[:len(formatMessages)-1]
  238. }
  239. }
  240. if fmtMessage.Content == nil && !(message.Role == "assistant" && message.ToolCalls != nil) {
  241. fmtMessage.SetStringContent("...")
  242. }
  243. formatMessages = append(formatMessages, fmtMessage)
  244. lastMessage = fmtMessage
  245. }
  246. claudeMessages := make([]dto.ClaudeMessage, 0)
  247. isFirstMessage := true
  248. // 初始化system消息数组,用于累积多个system消息
  249. var systemMessages []dto.ClaudeMediaMessage
  250. for _, message := range formatMessages {
  251. if message.Role == "system" {
  252. // 根据Claude API规范,system字段使用数组格式更有通用性
  253. if message.IsStringContent() {
  254. systemMessages = append(systemMessages, dto.ClaudeMediaMessage{
  255. Type: "text",
  256. Text: common.GetPointer[string](message.StringContent()),
  257. })
  258. } else {
  259. // 支持复合内容的system消息(虽然不常见,但需要考虑完整性)
  260. for _, ctx := range message.ParseContent() {
  261. if ctx.Type == "text" {
  262. systemMessages = append(systemMessages, dto.ClaudeMediaMessage{
  263. Type: "text",
  264. Text: common.GetPointer[string](ctx.Text),
  265. })
  266. }
  267. // 未来可以在这里扩展对图片等其他类型的支持
  268. }
  269. }
  270. } else {
  271. if isFirstMessage {
  272. isFirstMessage = false
  273. if message.Role != "user" {
  274. // fix: first message is assistant, add user message
  275. claudeMessage := dto.ClaudeMessage{
  276. Role: "user",
  277. Content: []dto.ClaudeMediaMessage{
  278. {
  279. Type: "text",
  280. Text: common.GetPointer[string]("..."),
  281. },
  282. },
  283. }
  284. claudeMessages = append(claudeMessages, claudeMessage)
  285. }
  286. }
  287. claudeMessage := dto.ClaudeMessage{
  288. Role: message.Role,
  289. }
  290. if message.Role == "tool" {
  291. if len(claudeMessages) > 0 && claudeMessages[len(claudeMessages)-1].Role == "user" {
  292. lastMessage := claudeMessages[len(claudeMessages)-1]
  293. if content, ok := lastMessage.Content.(string); ok {
  294. lastMessage.Content = []dto.ClaudeMediaMessage{
  295. {
  296. Type: "text",
  297. Text: common.GetPointer[string](content),
  298. },
  299. }
  300. }
  301. lastMessage.Content = append(lastMessage.Content.([]dto.ClaudeMediaMessage), dto.ClaudeMediaMessage{
  302. Type: "tool_result",
  303. ToolUseId: message.ToolCallId,
  304. Content: message.Content,
  305. })
  306. claudeMessages[len(claudeMessages)-1] = lastMessage
  307. continue
  308. } else {
  309. claudeMessage.Role = "user"
  310. claudeMessage.Content = []dto.ClaudeMediaMessage{
  311. {
  312. Type: "tool_result",
  313. ToolUseId: message.ToolCallId,
  314. Content: message.Content,
  315. },
  316. }
  317. }
  318. } else if message.IsStringContent() && message.ToolCalls == nil {
  319. claudeMessage.Content = message.StringContent()
  320. } else {
  321. claudeMediaMessages := make([]dto.ClaudeMediaMessage, 0)
  322. for _, mediaMessage := range message.ParseContent() {
  323. claudeMediaMessage := dto.ClaudeMediaMessage{
  324. Type: mediaMessage.Type,
  325. }
  326. if mediaMessage.Type == "text" {
  327. claudeMediaMessage.Text = common.GetPointer[string](mediaMessage.Text)
  328. } else {
  329. imageUrl := mediaMessage.GetImageMedia()
  330. claudeMediaMessage.Type = "image"
  331. claudeMediaMessage.Source = &dto.ClaudeMessageSource{
  332. Type: "base64",
  333. }
  334. // 使用统一的文件服务获取图片数据
  335. var source *types.FileSource
  336. if strings.HasPrefix(imageUrl.Url, "http") {
  337. source = types.NewURLFileSource(imageUrl.Url)
  338. } else {
  339. source = types.NewBase64FileSource(imageUrl.Url, "")
  340. }
  341. base64Data, mimeType, err := service.GetBase64Data(c, source, "formatting image for Claude")
  342. if err != nil {
  343. return nil, fmt.Errorf("get file data failed: %s", err.Error())
  344. }
  345. claudeMediaMessage.Source.MediaType = mimeType
  346. claudeMediaMessage.Source.Data = base64Data
  347. }
  348. claudeMediaMessages = append(claudeMediaMessages, claudeMediaMessage)
  349. }
  350. if message.ToolCalls != nil {
  351. for _, toolCall := range message.ParseToolCalls() {
  352. inputObj := make(map[string]any)
  353. if err := common.UnmarshalJsonStr(toolCall.Function.Arguments, &inputObj); err != nil {
  354. common.SysLog("tool call function arguments is not a map[string]any: " + fmt.Sprintf("%v", toolCall.Function.Arguments))
  355. inputObj = map[string]any{}
  356. }
  357. claudeMediaMessages = append(claudeMediaMessages, dto.ClaudeMediaMessage{
  358. Type: "tool_use",
  359. Id: toolCall.ID,
  360. Name: toolCall.Function.Name,
  361. Input: inputObj,
  362. })
  363. }
  364. }
  365. claudeMessage.Content = claudeMediaMessages
  366. }
  367. claudeMessages = append(claudeMessages, claudeMessage)
  368. }
  369. }
  370. // 设置累积的system消息
  371. if len(systemMessages) > 0 {
  372. claudeRequest.System = systemMessages
  373. }
  374. claudeRequest.Prompt = ""
  375. claudeRequest.Messages = claudeMessages
  376. return &claudeRequest, nil
  377. }
  378. func StreamResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.ChatCompletionsStreamResponse {
  379. var response dto.ChatCompletionsStreamResponse
  380. response.Object = "chat.completion.chunk"
  381. response.Model = claudeResponse.Model
  382. response.Choices = make([]dto.ChatCompletionsStreamResponseChoice, 0)
  383. tools := make([]dto.ToolCallResponse, 0)
  384. fcIdx := 0
  385. if claudeResponse.Index != nil {
  386. fcIdx = *claudeResponse.Index - 1
  387. if fcIdx < 0 {
  388. fcIdx = 0
  389. }
  390. }
  391. var choice dto.ChatCompletionsStreamResponseChoice
  392. if claudeResponse.Type == "message_start" {
  393. if claudeResponse.Message != nil {
  394. response.Id = claudeResponse.Message.Id
  395. response.Model = claudeResponse.Message.Model
  396. }
  397. //claudeUsage = &claudeResponse.Message.Usage
  398. choice.Delta.SetContentString("")
  399. choice.Delta.Role = "assistant"
  400. } else if claudeResponse.Type == "content_block_start" {
  401. if claudeResponse.ContentBlock != nil {
  402. // 如果是文本块,尽可能发送首段文本(若存在)
  403. if claudeResponse.ContentBlock.Type == "text" && claudeResponse.ContentBlock.Text != nil {
  404. choice.Delta.SetContentString(*claudeResponse.ContentBlock.Text)
  405. }
  406. if claudeResponse.ContentBlock.Type == "tool_use" {
  407. tools = append(tools, dto.ToolCallResponse{
  408. Index: common.GetPointer(fcIdx),
  409. ID: claudeResponse.ContentBlock.Id,
  410. Type: "function",
  411. Function: dto.FunctionResponse{
  412. Name: claudeResponse.ContentBlock.Name,
  413. Arguments: "",
  414. },
  415. })
  416. }
  417. } else {
  418. return nil
  419. }
  420. } else if claudeResponse.Type == "content_block_delta" {
  421. if claudeResponse.Delta != nil {
  422. choice.Delta.Content = claudeResponse.Delta.Text
  423. switch claudeResponse.Delta.Type {
  424. case "input_json_delta":
  425. tools = append(tools, dto.ToolCallResponse{
  426. Type: "function",
  427. Index: common.GetPointer(fcIdx),
  428. Function: dto.FunctionResponse{
  429. Arguments: *claudeResponse.Delta.PartialJson,
  430. },
  431. })
  432. case "signature_delta":
  433. // 加密的不处理
  434. signatureContent := "\n"
  435. choice.Delta.ReasoningContent = &signatureContent
  436. case "thinking_delta":
  437. choice.Delta.ReasoningContent = claudeResponse.Delta.Thinking
  438. }
  439. }
  440. } else if claudeResponse.Type == "message_delta" {
  441. if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
  442. finishReason := stopReasonClaude2OpenAI(*claudeResponse.Delta.StopReason)
  443. if finishReason != "null" {
  444. choice.FinishReason = &finishReason
  445. }
  446. }
  447. //claudeUsage = &claudeResponse.Usage
  448. } else if claudeResponse.Type == "message_stop" {
  449. return nil
  450. } else {
  451. return nil
  452. }
  453. if len(tools) > 0 {
  454. choice.Delta.Content = nil // compatible with other OpenAI derivative applications, like LobeOpenAICompatibleFactory ...
  455. choice.Delta.ToolCalls = tools
  456. }
  457. response.Choices = append(response.Choices, choice)
  458. return &response
  459. }
  460. func ResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.OpenAITextResponse {
  461. choices := make([]dto.OpenAITextResponseChoice, 0)
  462. fullTextResponse := dto.OpenAITextResponse{
  463. Id: fmt.Sprintf("chatcmpl-%s", common.GetUUID()),
  464. Object: "chat.completion",
  465. Created: common.GetTimestamp(),
  466. }
  467. var responseText string
  468. var responseThinking string
  469. if len(claudeResponse.Content) > 0 {
  470. responseText = claudeResponse.Content[0].GetText()
  471. if claudeResponse.Content[0].Thinking != nil {
  472. responseThinking = *claudeResponse.Content[0].Thinking
  473. }
  474. }
  475. tools := make([]dto.ToolCallResponse, 0)
  476. thinkingContent := ""
  477. fullTextResponse.Id = claudeResponse.Id
  478. for _, message := range claudeResponse.Content {
  479. switch message.Type {
  480. case "tool_use":
  481. args, _ := json.Marshal(message.Input)
  482. tools = append(tools, dto.ToolCallResponse{
  483. ID: message.Id,
  484. Type: "function", // compatible with other OpenAI derivative applications
  485. Function: dto.FunctionResponse{
  486. Name: message.Name,
  487. Arguments: string(args),
  488. },
  489. })
  490. case "thinking":
  491. // 加密的不管, 只输出明文的推理过程
  492. if message.Thinking != nil {
  493. thinkingContent = *message.Thinking
  494. }
  495. case "text":
  496. responseText = message.GetText()
  497. }
  498. }
  499. choice := dto.OpenAITextResponseChoice{
  500. Index: 0,
  501. Message: dto.Message{
  502. Role: "assistant",
  503. },
  504. FinishReason: stopReasonClaude2OpenAI(claudeResponse.StopReason),
  505. }
  506. choice.SetStringContent(responseText)
  507. if len(responseThinking) > 0 {
  508. choice.ReasoningContent = responseThinking
  509. }
  510. if len(tools) > 0 {
  511. choice.Message.SetToolCalls(tools)
  512. }
  513. choice.Message.ReasoningContent = thinkingContent
  514. fullTextResponse.Model = claudeResponse.Model
  515. choices = append(choices, choice)
  516. fullTextResponse.Choices = choices
  517. return &fullTextResponse
  518. }
  519. type ClaudeResponseInfo struct {
  520. ResponseId string
  521. Created int64
  522. Model string
  523. ResponseText strings.Builder
  524. Usage *dto.Usage
  525. Done bool
  526. }
  527. func buildMessageDeltaPatchUsage(claudeResponse *dto.ClaudeResponse, claudeInfo *ClaudeResponseInfo) *dto.ClaudeUsage {
  528. usage := &dto.ClaudeUsage{}
  529. if claudeResponse != nil && claudeResponse.Usage != nil {
  530. *usage = *claudeResponse.Usage
  531. }
  532. if claudeInfo == nil || claudeInfo.Usage == nil {
  533. return usage
  534. }
  535. if usage.InputTokens == 0 && claudeInfo.Usage.PromptTokens > 0 {
  536. usage.InputTokens = claudeInfo.Usage.PromptTokens
  537. }
  538. if usage.CacheReadInputTokens == 0 && claudeInfo.Usage.PromptTokensDetails.CachedTokens > 0 {
  539. usage.CacheReadInputTokens = claudeInfo.Usage.PromptTokensDetails.CachedTokens
  540. }
  541. if usage.CacheCreationInputTokens == 0 && claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens > 0 {
  542. usage.CacheCreationInputTokens = claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens
  543. }
  544. if usage.CacheCreation == nil && (claudeInfo.Usage.ClaudeCacheCreation5mTokens > 0 || claudeInfo.Usage.ClaudeCacheCreation1hTokens > 0) {
  545. usage.CacheCreation = &dto.ClaudeCacheCreationUsage{
  546. Ephemeral5mInputTokens: claudeInfo.Usage.ClaudeCacheCreation5mTokens,
  547. Ephemeral1hInputTokens: claudeInfo.Usage.ClaudeCacheCreation1hTokens,
  548. }
  549. }
  550. return usage
  551. }
  552. func shouldSkipClaudeMessageDeltaUsagePatch(info *relaycommon.RelayInfo) bool {
  553. if model_setting.GetGlobalSettings().PassThroughRequestEnabled {
  554. return true
  555. }
  556. if info == nil {
  557. return false
  558. }
  559. return info.ChannelSetting.PassThroughBodyEnabled
  560. }
  561. func patchClaudeMessageDeltaUsageData(data string, usage *dto.ClaudeUsage) string {
  562. if data == "" || usage == nil {
  563. return data
  564. }
  565. data = setMessageDeltaUsageInt(data, "usage.input_tokens", usage.InputTokens)
  566. data = setMessageDeltaUsageInt(data, "usage.cache_read_input_tokens", usage.CacheReadInputTokens)
  567. data = setMessageDeltaUsageInt(data, "usage.cache_creation_input_tokens", usage.CacheCreationInputTokens)
  568. if usage.CacheCreation != nil {
  569. data = setMessageDeltaUsageInt(data, "usage.cache_creation.ephemeral_5m_input_tokens", usage.CacheCreation.Ephemeral5mInputTokens)
  570. data = setMessageDeltaUsageInt(data, "usage.cache_creation.ephemeral_1h_input_tokens", usage.CacheCreation.Ephemeral1hInputTokens)
  571. }
  572. return data
  573. }
  574. func setMessageDeltaUsageInt(data string, path string, localValue int) string {
  575. if localValue <= 0 {
  576. return data
  577. }
  578. upstreamValue := gjson.Get(data, path)
  579. if upstreamValue.Exists() && upstreamValue.Int() > 0 {
  580. return data
  581. }
  582. patchedData, err := sjson.Set(data, path, localValue)
  583. if err != nil {
  584. return data
  585. }
  586. return patchedData
  587. }
  588. func FormatClaudeResponseInfo(claudeResponse *dto.ClaudeResponse, oaiResponse *dto.ChatCompletionsStreamResponse, claudeInfo *ClaudeResponseInfo) bool {
  589. if claudeInfo == nil {
  590. return false
  591. }
  592. if claudeInfo.Usage == nil {
  593. claudeInfo.Usage = &dto.Usage{}
  594. }
  595. if claudeResponse.Type == "message_start" {
  596. if claudeResponse.Message != nil {
  597. claudeInfo.ResponseId = claudeResponse.Message.Id
  598. claudeInfo.Model = claudeResponse.Message.Model
  599. }
  600. // message_start, 获取usage
  601. if claudeResponse.Message != nil && claudeResponse.Message.Usage != nil {
  602. claudeInfo.Usage.PromptTokens = claudeResponse.Message.Usage.InputTokens
  603. claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Message.Usage.CacheReadInputTokens
  604. claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Message.Usage.CacheCreationInputTokens
  605. claudeInfo.Usage.ClaudeCacheCreation5mTokens = claudeResponse.Message.Usage.GetCacheCreation5mTokens()
  606. claudeInfo.Usage.ClaudeCacheCreation1hTokens = claudeResponse.Message.Usage.GetCacheCreation1hTokens()
  607. claudeInfo.Usage.CompletionTokens = claudeResponse.Message.Usage.OutputTokens
  608. }
  609. } else if claudeResponse.Type == "content_block_delta" {
  610. if claudeResponse.Delta != nil {
  611. if claudeResponse.Delta.Text != nil {
  612. claudeInfo.ResponseText.WriteString(*claudeResponse.Delta.Text)
  613. }
  614. if claudeResponse.Delta.Thinking != nil {
  615. claudeInfo.ResponseText.WriteString(*claudeResponse.Delta.Thinking)
  616. }
  617. }
  618. } else if claudeResponse.Type == "message_delta" {
  619. // 最终的usage获取
  620. if claudeResponse.Usage != nil {
  621. if claudeResponse.Usage.InputTokens > 0 {
  622. // 不叠加,只取最新的
  623. claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
  624. }
  625. if claudeResponse.Usage.CacheReadInputTokens > 0 {
  626. claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Usage.CacheReadInputTokens
  627. }
  628. if claudeResponse.Usage.CacheCreationInputTokens > 0 {
  629. claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Usage.CacheCreationInputTokens
  630. }
  631. if cacheCreation5m := claudeResponse.Usage.GetCacheCreation5mTokens(); cacheCreation5m > 0 {
  632. claudeInfo.Usage.ClaudeCacheCreation5mTokens = cacheCreation5m
  633. }
  634. if cacheCreation1h := claudeResponse.Usage.GetCacheCreation1hTokens(); cacheCreation1h > 0 {
  635. claudeInfo.Usage.ClaudeCacheCreation1hTokens = cacheCreation1h
  636. }
  637. if claudeResponse.Usage.OutputTokens > 0 {
  638. claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
  639. }
  640. claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeInfo.Usage.CompletionTokens
  641. }
  642. // 判断是否完整
  643. claudeInfo.Done = true
  644. } else if claudeResponse.Type == "content_block_start" {
  645. } else {
  646. return false
  647. }
  648. if oaiResponse != nil {
  649. oaiResponse.Id = claudeInfo.ResponseId
  650. oaiResponse.Created = claudeInfo.Created
  651. oaiResponse.Model = claudeInfo.Model
  652. }
  653. return true
  654. }
  655. func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, data string) *types.NewAPIError {
  656. var claudeResponse dto.ClaudeResponse
  657. err := common.UnmarshalJsonStr(data, &claudeResponse)
  658. if err != nil {
  659. common.SysLog("error unmarshalling stream response: " + err.Error())
  660. return types.NewError(err, types.ErrorCodeBadResponseBody)
  661. }
  662. if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
  663. return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
  664. }
  665. if claudeResponse.StopReason != "" {
  666. maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
  667. }
  668. if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
  669. maybeMarkClaudeRefusal(c, *claudeResponse.Delta.StopReason)
  670. }
  671. if info.RelayFormat == types.RelayFormatClaude {
  672. FormatClaudeResponseInfo(&claudeResponse, nil, claudeInfo)
  673. if claudeResponse.Type == "message_start" {
  674. // message_start, 获取usage
  675. if claudeResponse.Message != nil {
  676. info.UpstreamModelName = claudeResponse.Message.Model
  677. }
  678. } else if claudeResponse.Type == "message_delta" {
  679. // 确保 message_delta 的 usage 包含完整的 input_tokens 和 cache 相关字段
  680. // 解决 AWS Bedrock 等上游返回的 message_delta 缺少这些字段的问题
  681. if !shouldSkipClaudeMessageDeltaUsagePatch(info) {
  682. data = patchClaudeMessageDeltaUsageData(data, buildMessageDeltaPatchUsage(&claudeResponse, claudeInfo))
  683. }
  684. }
  685. helper.ClaudeChunkData(c, claudeResponse, data)
  686. } else if info.RelayFormat == types.RelayFormatOpenAI {
  687. response := StreamResponseClaude2OpenAI(&claudeResponse)
  688. if !FormatClaudeResponseInfo(&claudeResponse, response, claudeInfo) {
  689. return nil
  690. }
  691. err = helper.ObjectData(c, response)
  692. if err != nil {
  693. logger.LogError(c, "send_stream_response_failed: "+err.Error())
  694. }
  695. }
  696. return nil
  697. }
  698. func HandleStreamFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo) {
  699. if claudeInfo.Usage.PromptTokens == 0 {
  700. //上游出错
  701. }
  702. if claudeInfo.Usage.CompletionTokens == 0 || !claudeInfo.Done {
  703. if common.DebugEnabled {
  704. common.SysLog("claude response usage is not complete, maybe upstream error")
  705. }
  706. claudeInfo.Usage = service.ResponseText2Usage(c, claudeInfo.ResponseText.String(), info.UpstreamModelName, claudeInfo.Usage.PromptTokens)
  707. }
  708. if info.RelayFormat == types.RelayFormatClaude {
  709. //
  710. } else if info.RelayFormat == types.RelayFormatOpenAI {
  711. if info.ShouldIncludeUsage {
  712. response := helper.GenerateFinalUsageResponse(claudeInfo.ResponseId, claudeInfo.Created, info.UpstreamModelName, *claudeInfo.Usage)
  713. err := helper.ObjectData(c, response)
  714. if err != nil {
  715. common.SysLog("send final response failed: " + err.Error())
  716. }
  717. }
  718. helper.Done(c)
  719. }
  720. }
  721. func ClaudeStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
  722. claudeInfo := &ClaudeResponseInfo{
  723. ResponseId: helper.GetResponseID(c),
  724. Created: common.GetTimestamp(),
  725. Model: info.UpstreamModelName,
  726. ResponseText: strings.Builder{},
  727. Usage: &dto.Usage{},
  728. }
  729. var err *types.NewAPIError
  730. helper.StreamScannerHandler(c, resp, info, func(data string) bool {
  731. err = HandleStreamResponseData(c, info, claudeInfo, data)
  732. if err != nil {
  733. return false
  734. }
  735. return true
  736. })
  737. if err != nil {
  738. return nil, err
  739. }
  740. HandleStreamFinalResponse(c, info, claudeInfo)
  741. return claudeInfo.Usage, nil
  742. }
  743. func HandleClaudeResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, httpResp *http.Response, data []byte) *types.NewAPIError {
  744. var claudeResponse dto.ClaudeResponse
  745. err := common.Unmarshal(data, &claudeResponse)
  746. if err != nil {
  747. return types.NewError(err, types.ErrorCodeBadResponseBody)
  748. }
  749. if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
  750. return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
  751. }
  752. maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
  753. if claudeInfo.Usage == nil {
  754. claudeInfo.Usage = &dto.Usage{}
  755. }
  756. if claudeResponse.Usage != nil {
  757. claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
  758. claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
  759. claudeInfo.Usage.TotalTokens = claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens
  760. claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Usage.CacheReadInputTokens
  761. claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Usage.CacheCreationInputTokens
  762. claudeInfo.Usage.ClaudeCacheCreation5mTokens = claudeResponse.Usage.GetCacheCreation5mTokens()
  763. claudeInfo.Usage.ClaudeCacheCreation1hTokens = claudeResponse.Usage.GetCacheCreation1hTokens()
  764. }
  765. var responseData []byte
  766. switch info.RelayFormat {
  767. case types.RelayFormatOpenAI:
  768. openaiResponse := ResponseClaude2OpenAI(&claudeResponse)
  769. openaiResponse.Usage = *claudeInfo.Usage
  770. responseData, err = json.Marshal(openaiResponse)
  771. if err != nil {
  772. return types.NewError(err, types.ErrorCodeBadResponseBody)
  773. }
  774. case types.RelayFormatClaude:
  775. responseData = data
  776. }
  777. if claudeResponse.Usage != nil && claudeResponse.Usage.ServerToolUse != nil && claudeResponse.Usage.ServerToolUse.WebSearchRequests > 0 {
  778. c.Set("claude_web_search_requests", claudeResponse.Usage.ServerToolUse.WebSearchRequests)
  779. }
  780. service.IOCopyBytesGracefully(c, httpResp, responseData)
  781. return nil
  782. }
  783. func ClaudeHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
  784. defer service.CloseResponseBodyGracefully(resp)
  785. claudeInfo := &ClaudeResponseInfo{
  786. ResponseId: helper.GetResponseID(c),
  787. Created: common.GetTimestamp(),
  788. Model: info.UpstreamModelName,
  789. ResponseText: strings.Builder{},
  790. Usage: &dto.Usage{},
  791. }
  792. responseBody, err := io.ReadAll(resp.Body)
  793. if err != nil {
  794. return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
  795. }
  796. if common.DebugEnabled {
  797. println("responseBody: ", string(responseBody))
  798. }
  799. handleErr := HandleClaudeResponseData(c, info, claudeInfo, resp, responseBody)
  800. if handleErr != nil {
  801. return nil, handleErr
  802. }
  803. return claudeInfo.Usage, nil
  804. }
  805. func mapToolChoice(toolChoice any, parallelToolCalls *bool) *dto.ClaudeToolChoice {
  806. var claudeToolChoice *dto.ClaudeToolChoice
  807. // 处理 tool_choice 字符串值
  808. if toolChoiceStr, ok := toolChoice.(string); ok {
  809. switch toolChoiceStr {
  810. case "auto":
  811. claudeToolChoice = &dto.ClaudeToolChoice{
  812. Type: "auto",
  813. }
  814. case "required":
  815. claudeToolChoice = &dto.ClaudeToolChoice{
  816. Type: "any",
  817. }
  818. case "none":
  819. claudeToolChoice = &dto.ClaudeToolChoice{
  820. Type: "none",
  821. }
  822. }
  823. } else if toolChoiceMap, ok := toolChoice.(map[string]interface{}); ok {
  824. // 处理 tool_choice 对象值
  825. if function, ok := toolChoiceMap["function"].(map[string]interface{}); ok {
  826. if toolName, ok := function["name"].(string); ok {
  827. claudeToolChoice = &dto.ClaudeToolChoice{
  828. Type: "tool",
  829. Name: toolName,
  830. }
  831. }
  832. }
  833. }
  834. // 处理 parallel_tool_calls
  835. if parallelToolCalls != nil {
  836. if claudeToolChoice == nil {
  837. // 如果没有 tool_choice,但有 parallel_tool_calls,创建默认的 auto 类型
  838. claudeToolChoice = &dto.ClaudeToolChoice{
  839. Type: "auto",
  840. }
  841. }
  842. // Anthropic schema: tool_choice.type=none does not accept extra fields.
  843. // When tools are disabled, parallel_tool_calls is irrelevant, so we drop it.
  844. if claudeToolChoice.Type != "none" {
  845. // 如果 parallel_tool_calls 为 true,则 disable_parallel_tool_use 为 false
  846. claudeToolChoice.DisableParallelToolUse = !*parallelToolCalls
  847. }
  848. }
  849. return claudeToolChoice
  850. }