You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 

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