Non puoi selezionare più di 25 argomenti Gli argomenti devono iniziare con una lettera o un numero, possono includere trattini ('-') e possono essere lunghi fino a 35 caratteri.
 
 
 

546 righe
15 KiB

  1. package openai
  2. import (
  3. "fmt"
  4. "io"
  5. "net/http"
  6. "strings"
  7. "time"
  8. "github.com/QuantumNous/new-api/common"
  9. "github.com/QuantumNous/new-api/dto"
  10. "github.com/QuantumNous/new-api/logger"
  11. relaycommon "github.com/QuantumNous/new-api/relay/common"
  12. "github.com/QuantumNous/new-api/relay/helper"
  13. "github.com/QuantumNous/new-api/service"
  14. "github.com/QuantumNous/new-api/types"
  15. "github.com/gin-gonic/gin"
  16. )
  17. func responsesStreamIndexKey(itemID string, idx *int) string {
  18. if itemID == "" {
  19. return ""
  20. }
  21. if idx == nil {
  22. return itemID
  23. }
  24. return fmt.Sprintf("%s:%d", itemID, *idx)
  25. }
  26. func stringDeltaFromPrefix(prev string, next string) string {
  27. if next == "" {
  28. return ""
  29. }
  30. if prev != "" && strings.HasPrefix(next, prev) {
  31. return next[len(prev):]
  32. }
  33. return next
  34. }
  35. func OaiResponsesToChatHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
  36. if resp == nil || resp.Body == nil {
  37. return nil, types.NewOpenAIError(fmt.Errorf("invalid response"), types.ErrorCodeBadResponse, http.StatusInternalServerError)
  38. }
  39. defer service.CloseResponseBodyGracefully(resp)
  40. var responsesResp dto.OpenAIResponsesResponse
  41. body, err := io.ReadAll(resp.Body)
  42. if err != nil {
  43. return nil, types.NewOpenAIError(err, types.ErrorCodeReadResponseBodyFailed, http.StatusInternalServerError)
  44. }
  45. if err := common.Unmarshal(body, &responsesResp); err != nil {
  46. apiErr := types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
  47. apiErr.UpstreamBody = service.TruncateBody(string(body))
  48. return nil, apiErr
  49. }
  50. if oaiError := responsesResp.GetOpenAIError(); oaiError != nil && oaiError.Type != "" {
  51. apiErr := types.WithOpenAIError(*oaiError, resp.StatusCode)
  52. apiErr.UpstreamBody = service.TruncateBody(string(body))
  53. return nil, apiErr
  54. }
  55. relaycommon.SetRelayChatID(c, responsesResp.ID)
  56. chatId := helper.GetResponseID(c)
  57. chatResp, usage, err := service.ResponsesResponseToChatCompletionsResponse(&responsesResp, chatId)
  58. if err != nil {
  59. apiErr := types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
  60. apiErr.UpstreamBody = service.TruncateBody(string(body))
  61. return nil, apiErr
  62. }
  63. if usage == nil || usage.TotalTokens == 0 {
  64. text := service.ExtractOutputTextFromResponses(&responsesResp)
  65. usage = service.ResponseText2Usage(c, text, info.UpstreamModelName, info.GetEstimatePromptTokens())
  66. chatResp.Usage = *usage
  67. }
  68. var responseBody []byte
  69. switch info.RelayFormat {
  70. case types.RelayFormatClaude:
  71. claudeResp := service.ResponseOpenAI2Claude(chatResp, info)
  72. responseBody, err = common.Marshal(claudeResp)
  73. case types.RelayFormatGemini:
  74. geminiResp := service.ResponseOpenAI2Gemini(chatResp, info)
  75. responseBody, err = common.Marshal(geminiResp)
  76. default:
  77. responseBody, err = common.Marshal(chatResp)
  78. }
  79. if err != nil {
  80. return nil, types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
  81. }
  82. service.IOCopyBytesGracefully(c, resp, responseBody)
  83. return usage, nil
  84. }
  85. func OaiResponsesToChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
  86. if resp == nil || resp.Body == nil {
  87. return nil, types.NewOpenAIError(fmt.Errorf("invalid response"), types.ErrorCodeBadResponse, http.StatusInternalServerError)
  88. }
  89. defer service.CloseResponseBodyGracefully(resp)
  90. responseId := helper.GetResponseID(c)
  91. createAt := time.Now().Unix()
  92. model := info.UpstreamModelName
  93. var (
  94. usage = &dto.Usage{}
  95. outputText strings.Builder
  96. usageText strings.Builder
  97. sentStart bool
  98. sentStop bool
  99. sawToolCall bool
  100. streamErr *types.NewAPIError
  101. )
  102. toolCallIndexByID := make(map[string]int)
  103. toolCallNameByID := make(map[string]string)
  104. toolCallArgsByID := make(map[string]string)
  105. toolCallNameSent := make(map[string]bool)
  106. toolCallCanonicalIDByItemID := make(map[string]string)
  107. hasSentReasoningSummary := false
  108. needsReasoningSummarySeparator := false
  109. //reasoningSummaryTextByKey := make(map[string]string)
  110. if info.RelayFormat == types.RelayFormatClaude && info.ClaudeConvertInfo == nil {
  111. info.ClaudeConvertInfo = &relaycommon.ClaudeConvertInfo{LastMessagesType: relaycommon.LastMessageTypeNone}
  112. }
  113. sendChatChunk := func(chunk *dto.ChatCompletionsStreamResponse) bool {
  114. if chunk == nil {
  115. return true
  116. }
  117. if info.RelayFormat == types.RelayFormatOpenAI {
  118. if err := helper.ObjectData(c, chunk); err != nil {
  119. streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
  120. return false
  121. }
  122. return true
  123. }
  124. chunkData, err := common.Marshal(chunk)
  125. if err != nil {
  126. streamErr = types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
  127. return false
  128. }
  129. if err := HandleStreamFormat(c, info, string(chunkData), false, false); err != nil {
  130. streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
  131. return false
  132. }
  133. return true
  134. }
  135. sendStartIfNeeded := func() bool {
  136. if sentStart {
  137. return true
  138. }
  139. if !sendChatChunk(helper.GenerateStartEmptyResponse(responseId, createAt, model, nil)) {
  140. return false
  141. }
  142. sentStart = true
  143. return true
  144. }
  145. //sendReasoningDelta := func(delta string) bool {
  146. // if delta == "" {
  147. // return true
  148. // }
  149. // if !sendStartIfNeeded() {
  150. // return false
  151. // }
  152. //
  153. // usageText.WriteString(delta)
  154. // chunk := &dto.ChatCompletionsStreamResponse{
  155. // Id: responseId,
  156. // Object: "chat.completion.chunk",
  157. // Created: createAt,
  158. // Model: model,
  159. // Choices: []dto.ChatCompletionsStreamResponseChoice{
  160. // {
  161. // Index: 0,
  162. // Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
  163. // ReasoningContent: &delta,
  164. // },
  165. // },
  166. // },
  167. // }
  168. // if err := helper.ObjectData(c, chunk); err != nil {
  169. // streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
  170. // return false
  171. // }
  172. // return true
  173. //}
  174. sendReasoningSummaryDelta := func(delta string) bool {
  175. if delta == "" {
  176. return true
  177. }
  178. if needsReasoningSummarySeparator {
  179. if strings.HasPrefix(delta, "\n\n") {
  180. needsReasoningSummarySeparator = false
  181. } else if strings.HasPrefix(delta, "\n") {
  182. delta = "\n" + delta
  183. needsReasoningSummarySeparator = false
  184. } else {
  185. delta = "\n\n" + delta
  186. needsReasoningSummarySeparator = false
  187. }
  188. }
  189. if !sendStartIfNeeded() {
  190. return false
  191. }
  192. usageText.WriteString(delta)
  193. chunk := &dto.ChatCompletionsStreamResponse{
  194. Id: responseId,
  195. Object: "chat.completion.chunk",
  196. Created: createAt,
  197. Model: model,
  198. Choices: []dto.ChatCompletionsStreamResponseChoice{
  199. {
  200. Index: 0,
  201. Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
  202. ReasoningContent: &delta,
  203. },
  204. },
  205. },
  206. }
  207. if !sendChatChunk(chunk) {
  208. return false
  209. }
  210. hasSentReasoningSummary = true
  211. return true
  212. }
  213. sendToolCallDelta := func(callID string, name string, argsDelta string) bool {
  214. if callID == "" {
  215. return true
  216. }
  217. if outputText.Len() > 0 {
  218. // Prefer streaming assistant text over tool calls to match non-stream behavior.
  219. return true
  220. }
  221. if !sendStartIfNeeded() {
  222. return false
  223. }
  224. idx, ok := toolCallIndexByID[callID]
  225. if !ok {
  226. idx = len(toolCallIndexByID)
  227. toolCallIndexByID[callID] = idx
  228. }
  229. if name != "" {
  230. toolCallNameByID[callID] = name
  231. }
  232. if toolCallNameByID[callID] != "" {
  233. name = toolCallNameByID[callID]
  234. }
  235. tool := dto.ToolCallResponse{
  236. ID: callID,
  237. Type: "function",
  238. Function: dto.FunctionResponse{
  239. Arguments: argsDelta,
  240. },
  241. }
  242. tool.SetIndex(idx)
  243. if name != "" && !toolCallNameSent[callID] {
  244. tool.Function.Name = name
  245. toolCallNameSent[callID] = true
  246. }
  247. chunk := &dto.ChatCompletionsStreamResponse{
  248. Id: responseId,
  249. Object: "chat.completion.chunk",
  250. Created: createAt,
  251. Model: model,
  252. Choices: []dto.ChatCompletionsStreamResponseChoice{
  253. {
  254. Index: 0,
  255. Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
  256. ToolCalls: []dto.ToolCallResponse{tool},
  257. },
  258. },
  259. },
  260. }
  261. if !sendChatChunk(chunk) {
  262. return false
  263. }
  264. sawToolCall = true
  265. // Include tool call data in the local builder for fallback token estimation.
  266. if tool.Function.Name != "" {
  267. usageText.WriteString(tool.Function.Name)
  268. }
  269. if argsDelta != "" {
  270. usageText.WriteString(argsDelta)
  271. }
  272. return true
  273. }
  274. helper.StreamScannerHandler(c, resp, info, func(data string) bool {
  275. if streamErr != nil {
  276. return false
  277. }
  278. var streamResp dto.ResponsesStreamResponse
  279. if err := common.UnmarshalJsonStr(data, &streamResp); err != nil {
  280. logger.LogError(c, "failed to unmarshal responses stream event: "+err.Error())
  281. return true
  282. }
  283. switch streamResp.Type {
  284. case "response.created":
  285. if streamResp.Response != nil {
  286. if streamResp.Response.Model != "" {
  287. model = streamResp.Response.Model
  288. }
  289. if streamResp.Response.CreatedAt != 0 {
  290. createAt = int64(streamResp.Response.CreatedAt)
  291. }
  292. }
  293. //case "response.reasoning_text.delta":
  294. //if !sendReasoningDelta(streamResp.Delta) {
  295. // return false
  296. //}
  297. //case "response.reasoning_text.done":
  298. case "response.reasoning_summary_text.delta":
  299. if !sendReasoningSummaryDelta(streamResp.Delta) {
  300. return false
  301. }
  302. case "response.reasoning_summary_text.done":
  303. if hasSentReasoningSummary {
  304. needsReasoningSummarySeparator = true
  305. }
  306. //case "response.reasoning_summary_part.added", "response.reasoning_summary_part.done":
  307. // key := responsesStreamIndexKey(strings.TrimSpace(streamResp.ItemID), streamResp.SummaryIndex)
  308. // if key == "" || streamResp.Part == nil {
  309. // break
  310. // }
  311. // // Only handle summary text parts, ignore other part types.
  312. // if streamResp.Part.Type != "" && streamResp.Part.Type != "summary_text" {
  313. // break
  314. // }
  315. // prev := reasoningSummaryTextByKey[key]
  316. // next := streamResp.Part.Text
  317. // delta := stringDeltaFromPrefix(prev, next)
  318. // reasoningSummaryTextByKey[key] = next
  319. // if !sendReasoningSummaryDelta(delta) {
  320. // return false
  321. // }
  322. case "response.output_text.delta":
  323. if !sendStartIfNeeded() {
  324. return false
  325. }
  326. if streamResp.Delta != "" {
  327. outputText.WriteString(streamResp.Delta)
  328. usageText.WriteString(streamResp.Delta)
  329. delta := streamResp.Delta
  330. chunk := &dto.ChatCompletionsStreamResponse{
  331. Id: responseId,
  332. Object: "chat.completion.chunk",
  333. Created: createAt,
  334. Model: model,
  335. Choices: []dto.ChatCompletionsStreamResponseChoice{
  336. {
  337. Index: 0,
  338. Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
  339. Content: &delta,
  340. },
  341. },
  342. },
  343. }
  344. if !sendChatChunk(chunk) {
  345. return false
  346. }
  347. }
  348. case "response.output_item.added", "response.output_item.done":
  349. if streamResp.Item == nil {
  350. break
  351. }
  352. if streamResp.Item.Type != "function_call" {
  353. break
  354. }
  355. itemID := strings.TrimSpace(streamResp.Item.ID)
  356. callID := strings.TrimSpace(streamResp.Item.CallId)
  357. if callID == "" {
  358. callID = itemID
  359. }
  360. if itemID != "" && callID != "" {
  361. toolCallCanonicalIDByItemID[itemID] = callID
  362. }
  363. name := strings.TrimSpace(streamResp.Item.Name)
  364. if name != "" {
  365. toolCallNameByID[callID] = name
  366. }
  367. newArgs := streamResp.Item.Arguments
  368. prevArgs := toolCallArgsByID[callID]
  369. argsDelta := ""
  370. if newArgs != "" {
  371. if strings.HasPrefix(newArgs, prevArgs) {
  372. argsDelta = newArgs[len(prevArgs):]
  373. } else {
  374. argsDelta = newArgs
  375. }
  376. toolCallArgsByID[callID] = newArgs
  377. }
  378. if !sendToolCallDelta(callID, name, argsDelta) {
  379. return false
  380. }
  381. case "response.function_call_arguments.delta":
  382. itemID := strings.TrimSpace(streamResp.ItemID)
  383. callID := toolCallCanonicalIDByItemID[itemID]
  384. if callID == "" {
  385. callID = itemID
  386. }
  387. if callID == "" {
  388. break
  389. }
  390. toolCallArgsByID[callID] += streamResp.Delta
  391. if !sendToolCallDelta(callID, "", streamResp.Delta) {
  392. return false
  393. }
  394. case "response.function_call_arguments.done":
  395. case "response.completed":
  396. if streamResp.Response != nil {
  397. if streamResp.Response.ID != "" {
  398. responseId = streamResp.Response.ID
  399. relaycommon.SetRelayChatID(c, responseId)
  400. }
  401. if streamResp.Response.Model != "" {
  402. model = streamResp.Response.Model
  403. }
  404. if streamResp.Response.CreatedAt != 0 {
  405. createAt = int64(streamResp.Response.CreatedAt)
  406. }
  407. if streamResp.Response.Usage != nil {
  408. if streamResp.Response.Usage.InputTokens != 0 {
  409. usage.PromptTokens = streamResp.Response.Usage.InputTokens
  410. usage.InputTokens = streamResp.Response.Usage.InputTokens
  411. }
  412. if streamResp.Response.Usage.OutputTokens != 0 {
  413. usage.CompletionTokens = streamResp.Response.Usage.OutputTokens
  414. usage.OutputTokens = streamResp.Response.Usage.OutputTokens
  415. }
  416. if streamResp.Response.Usage.TotalTokens != 0 {
  417. usage.TotalTokens = streamResp.Response.Usage.TotalTokens
  418. } else {
  419. usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
  420. }
  421. if streamResp.Response.Usage.InputTokensDetails != nil {
  422. usage.PromptTokensDetails.CachedTokens = streamResp.Response.Usage.InputTokensDetails.CachedTokens
  423. usage.PromptTokensDetails.ImageTokens = streamResp.Response.Usage.InputTokensDetails.ImageTokens
  424. usage.PromptTokensDetails.AudioTokens = streamResp.Response.Usage.InputTokensDetails.AudioTokens
  425. }
  426. if streamResp.Response.Usage.CompletionTokenDetails.ReasoningTokens != 0 {
  427. usage.CompletionTokenDetails.ReasoningTokens = streamResp.Response.Usage.CompletionTokenDetails.ReasoningTokens
  428. }
  429. }
  430. }
  431. if !sendStartIfNeeded() {
  432. return false
  433. }
  434. if !sentStop {
  435. if info.RelayFormat == types.RelayFormatClaude && info.ClaudeConvertInfo != nil {
  436. info.ClaudeConvertInfo.Usage = usage
  437. }
  438. finishReason := "stop"
  439. if sawToolCall && outputText.Len() == 0 {
  440. finishReason = "tool_calls"
  441. }
  442. stop := helper.GenerateStopResponse(responseId, createAt, model, finishReason)
  443. if !sendChatChunk(stop) {
  444. return false
  445. }
  446. sentStop = true
  447. }
  448. case "response.error", "response.failed", "error":
  449. streamErr = handleResponsesStreamError(streamResp, data)
  450. return false
  451. default:
  452. }
  453. return true
  454. })
  455. if streamErr != nil {
  456. return nil, streamErr
  457. }
  458. if usage.TotalTokens == 0 {
  459. usage = service.ResponseText2Usage(c, usageText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
  460. }
  461. if !sentStart {
  462. if !sendChatChunk(helper.GenerateStartEmptyResponse(responseId, createAt, model, nil)) {
  463. return nil, streamErr
  464. }
  465. }
  466. if !sentStop {
  467. if info.RelayFormat == types.RelayFormatClaude && info.ClaudeConvertInfo != nil {
  468. info.ClaudeConvertInfo.Usage = usage
  469. }
  470. finishReason := "stop"
  471. if sawToolCall && outputText.Len() == 0 {
  472. finishReason = "tool_calls"
  473. }
  474. stop := helper.GenerateStopResponse(responseId, createAt, model, finishReason)
  475. if !sendChatChunk(stop) {
  476. return nil, streamErr
  477. }
  478. }
  479. if info.RelayFormat == types.RelayFormatOpenAI && info.ShouldIncludeUsage && usage != nil {
  480. if err := helper.ObjectData(c, helper.GenerateFinalUsageResponse(responseId, createAt, model, *usage)); err != nil {
  481. return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
  482. }
  483. }
  484. if info.RelayFormat == types.RelayFormatOpenAI {
  485. helper.Done(c)
  486. }
  487. return usage, nil
  488. }