adaptor.go 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420
  1. package vertex
  2. import (
  3. "encoding/json"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "net/http"
  8. "strings"
  9. "github.com/QuantumNous/new-api/common"
  10. "github.com/QuantumNous/new-api/dto"
  11. "github.com/QuantumNous/new-api/relay/channel"
  12. "github.com/QuantumNous/new-api/relay/channel/claude"
  13. "github.com/QuantumNous/new-api/relay/channel/gemini"
  14. "github.com/QuantumNous/new-api/relay/channel/openai"
  15. relaycommon "github.com/QuantumNous/new-api/relay/common"
  16. "github.com/QuantumNous/new-api/relay/constant"
  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. )
  22. const (
  23. RequestModeClaude = 1
  24. RequestModeGemini = 2
  25. RequestModeOpenSource = 3
  26. )
  27. var claudeModelMap = map[string]string{
  28. "claude-3-sonnet-20240229": "claude-3-sonnet@20240229",
  29. "claude-3-opus-20240229": "claude-3-opus@20240229",
  30. "claude-3-haiku-20240307": "claude-3-haiku@20240307",
  31. "claude-3-5-sonnet-20240620": "claude-3-5-sonnet@20240620",
  32. "claude-3-5-sonnet-20241022": "claude-3-5-sonnet-v2@20241022",
  33. "claude-3-7-sonnet-20250219": "claude-3-7-sonnet@20250219",
  34. "claude-sonnet-4-20250514": "claude-sonnet-4@20250514",
  35. "claude-opus-4-20250514": "claude-opus-4@20250514",
  36. "claude-opus-4-1-20250805": "claude-opus-4-1@20250805",
  37. "claude-sonnet-4-5-20250929": "claude-sonnet-4-5@20250929",
  38. "claude-haiku-4-5-20251001": "claude-haiku-4-5@20251001",
  39. "claude-opus-4-5-20251101": "claude-opus-4-5@20251101",
  40. "claude-opus-4-6": "claude-opus-4-6",
  41. }
  42. const anthropicVersion = "vertex-2023-10-16"
  43. type Adaptor struct {
  44. RequestMode int
  45. AccountCredentials Credentials
  46. }
  47. func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
  48. // Vertex AI does not support functionResponse.id; keep it stripped here for consistency.
  49. if model_setting.GetGeminiSettings().RemoveFunctionResponseIdEnabled {
  50. removeFunctionResponseID(request)
  51. }
  52. geminiAdaptor := gemini.Adaptor{}
  53. return geminiAdaptor.ConvertGeminiRequest(c, info, request)
  54. }
  55. func removeFunctionResponseID(request *dto.GeminiChatRequest) {
  56. if request == nil {
  57. return
  58. }
  59. if len(request.Contents) > 0 {
  60. for i := range request.Contents {
  61. if len(request.Contents[i].Parts) == 0 {
  62. continue
  63. }
  64. for j := range request.Contents[i].Parts {
  65. part := &request.Contents[i].Parts[j]
  66. if part.FunctionResponse == nil {
  67. continue
  68. }
  69. if len(part.FunctionResponse.ID) > 0 {
  70. part.FunctionResponse.ID = nil
  71. }
  72. }
  73. }
  74. }
  75. if len(request.Requests) > 0 {
  76. for i := range request.Requests {
  77. removeFunctionResponseID(&request.Requests[i])
  78. }
  79. }
  80. }
  81. func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
  82. if v, ok := claudeModelMap[info.UpstreamModelName]; ok {
  83. c.Set("request_model", v)
  84. } else {
  85. c.Set("request_model", request.Model)
  86. }
  87. vertexClaudeReq := copyRequest(request, anthropicVersion)
  88. return vertexClaudeReq, nil
  89. }
  90. func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.AudioRequest) (io.Reader, error) {
  91. //TODO implement me
  92. return nil, errors.New("not implemented")
  93. }
  94. func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.ImageRequest) (any, error) {
  95. geminiAdaptor := gemini.Adaptor{}
  96. return geminiAdaptor.ConvertImageRequest(c, info, request)
  97. }
  98. func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
  99. if strings.HasPrefix(info.UpstreamModelName, "claude") {
  100. a.RequestMode = RequestModeClaude
  101. } else if strings.Contains(info.UpstreamModelName, "llama") ||
  102. // open source models
  103. strings.Contains(info.UpstreamModelName, "-maas") {
  104. a.RequestMode = RequestModeOpenSource
  105. } else {
  106. a.RequestMode = RequestModeGemini
  107. }
  108. }
  109. func (a *Adaptor) getRequestUrl(info *relaycommon.RelayInfo, modelName, suffix string) (string, error) {
  110. region := GetModelRegion(info.ApiVersion, info.OriginModelName)
  111. if info.ChannelOtherSettings.VertexKeyType != dto.VertexKeyTypeAPIKey {
  112. adc := &Credentials{}
  113. if err := common.Unmarshal([]byte(info.ApiKey), adc); err != nil {
  114. return "", fmt.Errorf("failed to decode credentials file: %w", err)
  115. }
  116. a.AccountCredentials = *adc
  117. if a.RequestMode == RequestModeGemini {
  118. if region == "global" {
  119. return fmt.Sprintf(
  120. "https://aiplatform.googleapis.com/v1/projects/%s/locations/global/publishers/google/models/%s:%s",
  121. adc.ProjectID,
  122. modelName,
  123. suffix,
  124. ), nil
  125. } else {
  126. return fmt.Sprintf(
  127. "https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/google/models/%s:%s",
  128. region,
  129. adc.ProjectID,
  130. region,
  131. modelName,
  132. suffix,
  133. ), nil
  134. }
  135. } else if a.RequestMode == RequestModeClaude {
  136. if region == "global" {
  137. return fmt.Sprintf(
  138. "https://aiplatform.googleapis.com/v1/projects/%s/locations/global/publishers/anthropic/models/%s:%s",
  139. adc.ProjectID,
  140. modelName,
  141. suffix,
  142. ), nil
  143. } else {
  144. return fmt.Sprintf(
  145. "https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/anthropic/models/%s:%s",
  146. region,
  147. adc.ProjectID,
  148. region,
  149. modelName,
  150. suffix,
  151. ), nil
  152. }
  153. } else if a.RequestMode == RequestModeOpenSource {
  154. return fmt.Sprintf(
  155. "https://aiplatform.googleapis.com/v1beta1/projects/%s/locations/%s/endpoints/openapi/chat/completions",
  156. adc.ProjectID,
  157. region,
  158. ), nil
  159. }
  160. } else {
  161. var keyPrefix string
  162. if strings.HasSuffix(suffix, "?alt=sse") {
  163. keyPrefix = "&"
  164. } else {
  165. keyPrefix = "?"
  166. }
  167. if region == "global" {
  168. return fmt.Sprintf(
  169. "https://aiplatform.googleapis.com/v1/publishers/google/models/%s:%s%skey=%s",
  170. modelName,
  171. suffix,
  172. keyPrefix,
  173. info.ApiKey,
  174. ), nil
  175. } else {
  176. return fmt.Sprintf(
  177. "https://%s-aiplatform.googleapis.com/v1/publishers/google/models/%s:%s%skey=%s",
  178. region,
  179. modelName,
  180. suffix,
  181. keyPrefix,
  182. info.ApiKey,
  183. ), nil
  184. }
  185. }
  186. return "", errors.New("unsupported request mode")
  187. }
  188. func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
  189. suffix := ""
  190. if a.RequestMode == RequestModeGemini {
  191. if model_setting.GetGeminiSettings().ThinkingAdapterEnabled &&
  192. !model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) {
  193. // 新增逻辑:处理 -thinking-<budget> 格式
  194. if strings.Contains(info.UpstreamModelName, "-thinking-") {
  195. parts := strings.Split(info.UpstreamModelName, "-thinking-")
  196. info.UpstreamModelName = parts[0]
  197. } else if strings.HasSuffix(info.UpstreamModelName, "-thinking") { // 旧的适配
  198. info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-thinking")
  199. } else if strings.HasSuffix(info.UpstreamModelName, "-nothinking") {
  200. info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-nothinking")
  201. } else if baseModel, level, ok := reasoning.TrimEffortSuffix(info.UpstreamModelName); ok && level != "" {
  202. info.UpstreamModelName = baseModel
  203. }
  204. }
  205. if info.IsStream {
  206. suffix = "streamGenerateContent?alt=sse"
  207. } else {
  208. suffix = "generateContent"
  209. }
  210. if strings.HasPrefix(info.UpstreamModelName, "imagen") {
  211. suffix = "predict"
  212. }
  213. return a.getRequestUrl(info, info.UpstreamModelName, suffix)
  214. } else if a.RequestMode == RequestModeClaude {
  215. if info.IsStream {
  216. suffix = "streamRawPredict?alt=sse"
  217. } else {
  218. suffix = "rawPredict"
  219. }
  220. model := info.UpstreamModelName
  221. if v, ok := claudeModelMap[info.UpstreamModelName]; ok {
  222. model = v
  223. }
  224. return a.getRequestUrl(info, model, suffix)
  225. } else if a.RequestMode == RequestModeOpenSource {
  226. return a.getRequestUrl(info, "", "")
  227. }
  228. return "", errors.New("unsupported request mode")
  229. }
  230. func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) error {
  231. channel.SetupApiRequestHeader(info, c, req)
  232. if info.ChannelOtherSettings.VertexKeyType != dto.VertexKeyTypeAPIKey {
  233. accessToken, err := getAccessToken(a, info)
  234. if err != nil {
  235. return err
  236. }
  237. req.Set("Authorization", "Bearer "+accessToken)
  238. }
  239. if a.AccountCredentials.ProjectID != "" {
  240. req.Set("x-goog-user-project", a.AccountCredentials.ProjectID)
  241. }
  242. if strings.Contains(info.UpstreamModelName, "claude") {
  243. claude.CommonClaudeHeadersOperation(c, req, info)
  244. }
  245. return nil
  246. }
  247. func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeneralOpenAIRequest) (any, error) {
  248. if request == nil {
  249. return nil, errors.New("request is nil")
  250. }
  251. if a.RequestMode == RequestModeGemini && strings.HasPrefix(info.UpstreamModelName, "imagen") {
  252. prompt := ""
  253. for _, m := range request.Messages {
  254. if m.Role == "user" {
  255. prompt = m.StringContent()
  256. if prompt != "" {
  257. break
  258. }
  259. }
  260. }
  261. if prompt == "" {
  262. if p, ok := request.Prompt.(string); ok {
  263. prompt = p
  264. }
  265. }
  266. if prompt == "" {
  267. return nil, errors.New("prompt is required for image generation")
  268. }
  269. imgReq := dto.ImageRequest{
  270. Model: request.Model,
  271. Prompt: prompt,
  272. N: 1,
  273. Size: "1024x1024",
  274. }
  275. if request.N > 0 {
  276. imgReq.N = uint(request.N)
  277. }
  278. if request.Size != "" {
  279. imgReq.Size = request.Size
  280. }
  281. if len(request.ExtraBody) > 0 {
  282. var extra map[string]any
  283. if err := json.Unmarshal(request.ExtraBody, &extra); err == nil {
  284. if n, ok := extra["n"].(float64); ok && n > 0 {
  285. imgReq.N = uint(n)
  286. }
  287. if size, ok := extra["size"].(string); ok {
  288. imgReq.Size = size
  289. }
  290. // accept aspectRatio in extra body (top-level or under parameters)
  291. if ar, ok := extra["aspectRatio"].(string); ok && ar != "" {
  292. imgReq.Size = ar
  293. }
  294. if params, ok := extra["parameters"].(map[string]any); ok {
  295. if ar, ok := params["aspectRatio"].(string); ok && ar != "" {
  296. imgReq.Size = ar
  297. }
  298. }
  299. }
  300. }
  301. c.Set("request_model", request.Model)
  302. return a.ConvertImageRequest(c, info, imgReq)
  303. }
  304. if a.RequestMode == RequestModeClaude {
  305. claudeReq, err := claude.RequestOpenAI2ClaudeMessage(c, *request)
  306. if err != nil {
  307. return nil, err
  308. }
  309. vertexClaudeReq := copyRequest(claudeReq, anthropicVersion)
  310. c.Set("request_model", claudeReq.Model)
  311. info.UpstreamModelName = claudeReq.Model
  312. return vertexClaudeReq, nil
  313. } else if a.RequestMode == RequestModeGemini {
  314. geminiRequest, err := gemini.CovertOpenAI2Gemini(c, *request, info)
  315. if err != nil {
  316. return nil, err
  317. }
  318. c.Set("request_model", request.Model)
  319. return geminiRequest, nil
  320. } else if a.RequestMode == RequestModeOpenSource {
  321. return request, nil
  322. }
  323. return nil, errors.New("unsupported request mode")
  324. }
  325. func (a *Adaptor) ConvertRerankRequest(c *gin.Context, relayMode int, request dto.RerankRequest) (any, error) {
  326. return nil, nil
  327. }
  328. func (a *Adaptor) ConvertEmbeddingRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.EmbeddingRequest) (any, error) {
  329. //TODO implement me
  330. return nil, errors.New("not implemented")
  331. }
  332. func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
  333. // TODO implement me
  334. return nil, errors.New("not implemented")
  335. }
  336. func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, requestBody io.Reader) (any, error) {
  337. return channel.DoApiRequest(a, c, info, requestBody)
  338. }
  339. func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage any, err *types.NewAPIError) {
  340. if info.IsStream {
  341. switch a.RequestMode {
  342. case RequestModeClaude:
  343. return claude.ClaudeStreamHandler(c, resp, info)
  344. case RequestModeGemini:
  345. if info.RelayMode == constant.RelayModeGemini {
  346. return gemini.GeminiTextGenerationStreamHandler(c, info, resp)
  347. } else {
  348. return gemini.GeminiChatStreamHandler(c, info, resp)
  349. }
  350. case RequestModeOpenSource:
  351. return openai.OaiStreamHandler(c, info, resp)
  352. }
  353. } else {
  354. switch a.RequestMode {
  355. case RequestModeClaude:
  356. return claude.ClaudeHandler(c, resp, info)
  357. case RequestModeGemini:
  358. if info.RelayMode == constant.RelayModeGemini {
  359. return gemini.GeminiTextGenerationHandler(c, info, resp)
  360. } else {
  361. if strings.HasPrefix(info.UpstreamModelName, "imagen") {
  362. return gemini.GeminiImageHandler(c, info, resp)
  363. }
  364. return gemini.GeminiChatHandler(c, info, resp)
  365. }
  366. case RequestModeOpenSource:
  367. return openai.OpenaiHandler(c, info, resp)
  368. }
  369. }
  370. return
  371. }
  372. func (a *Adaptor) GetModelList() []string {
  373. var modelList []string
  374. for i, s := range ModelList {
  375. modelList = append(modelList, s)
  376. ModelList[i] = s
  377. }
  378. for i, s := range claude.ModelList {
  379. modelList = append(modelList, s)
  380. claude.ModelList[i] = s
  381. }
  382. for i, s := range gemini.ModelList {
  383. modelList = append(modelList, s)
  384. gemini.ModelList[i] = s
  385. }
  386. return modelList
  387. }
  388. func (a *Adaptor) GetChannelName() string {
  389. return ChannelName
  390. }