relay.go 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254
  1. package controller
  2. import (
  3. "bufio"
  4. "bytes"
  5. "encoding/json"
  6. "fmt"
  7. "github.com/gin-gonic/gin"
  8. "github.com/pkoukk/tiktoken-go"
  9. "io"
  10. "net/http"
  11. "one-api/common"
  12. "one-api/model"
  13. "strings"
  14. )
  15. type Message struct {
  16. Role string `json:"role"`
  17. Content string `json:"content"`
  18. }
  19. type TextRequest struct {
  20. Model string `json:"model"`
  21. Messages []Message `json:"messages"`
  22. Prompt string `json:"prompt"`
  23. //Stream bool `json:"stream"`
  24. }
  25. type Usage struct {
  26. PromptTokens int `json:"prompt_tokens"`
  27. CompletionTokens int `json:"completion_tokens"`
  28. TotalTokens int `json:"total_tokens"`
  29. }
  30. type TextResponse struct {
  31. Usage `json:"usage"`
  32. }
  33. type StreamResponse struct {
  34. Choices []struct {
  35. Delta struct {
  36. Content string `json:"content"`
  37. } `json:"delta"`
  38. FinishReason string `json:"finish_reason"`
  39. } `json:"choices"`
  40. }
  41. var tokenEncoder, _ = tiktoken.GetEncoding("cl100k_base")
  42. func countToken(text string) int {
  43. token := tokenEncoder.Encode(text, nil, nil)
  44. return len(token)
  45. }
  46. func Relay(c *gin.Context) {
  47. err := relayHelper(c)
  48. if err != nil {
  49. c.JSON(http.StatusOK, gin.H{
  50. "error": gin.H{
  51. "message": err.Error(),
  52. "type": "one_api_error",
  53. },
  54. })
  55. }
  56. }
  57. func relayHelper(c *gin.Context) error {
  58. channelType := c.GetInt("channel")
  59. tokenId := c.GetInt("token_id")
  60. consumeQuota := c.GetBool("consume_quota")
  61. var textRequest TextRequest
  62. if consumeQuota || channelType == common.ChannelTypeAzure {
  63. requestBody, err := io.ReadAll(c.Request.Body)
  64. if err != nil {
  65. return err
  66. }
  67. err = c.Request.Body.Close()
  68. if err != nil {
  69. return err
  70. }
  71. err = json.Unmarshal(requestBody, &textRequest)
  72. if err != nil {
  73. return err
  74. }
  75. // Reset request body
  76. c.Request.Body = io.NopCloser(bytes.NewBuffer(requestBody))
  77. }
  78. baseURL := common.ChannelBaseURLs[channelType]
  79. requestURL := c.Request.URL.String()
  80. if channelType == common.ChannelTypeCustom {
  81. baseURL = c.GetString("base_url")
  82. }
  83. fullRequestURL := fmt.Sprintf("%s%s", baseURL, requestURL)
  84. if channelType == common.ChannelTypeAzure {
  85. // https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
  86. baseURL = c.GetString("base_url")
  87. task := strings.TrimPrefix(requestURL, "/v1/")
  88. model_ := textRequest.Model
  89. fullRequestURL = fmt.Sprintf("%s/openai/deployments/%s/%s", baseURL, model_, task)
  90. }
  91. req, err := http.NewRequest(c.Request.Method, fullRequestURL, c.Request.Body)
  92. if err != nil {
  93. return err
  94. }
  95. if channelType == common.ChannelTypeAzure {
  96. key := c.Request.Header.Get("Authorization")
  97. key = strings.TrimPrefix(key, "Bearer ")
  98. req.Header.Set("api-key", key)
  99. } else {
  100. req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
  101. }
  102. req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
  103. req.Header.Set("Accept", c.Request.Header.Get("Accept"))
  104. req.Header.Set("Connection", c.Request.Header.Get("Connection"))
  105. client := &http.Client{}
  106. resp, err := client.Do(req)
  107. if err != nil {
  108. return err
  109. }
  110. err = req.Body.Close()
  111. if err != nil {
  112. return err
  113. }
  114. err = c.Request.Body.Close()
  115. if err != nil {
  116. return err
  117. }
  118. var textResponse TextResponse
  119. isStream := resp.Header.Get("Content-Type") == "text/event-stream"
  120. var streamResponseText string
  121. defer func() {
  122. if consumeQuota {
  123. quota := 0
  124. usingGPT4 := strings.HasPrefix(textRequest.Model, "gpt-4")
  125. completionRatio := 1
  126. if usingGPT4 {
  127. completionRatio = 2
  128. }
  129. if isStream {
  130. var promptText string
  131. for _, message := range textRequest.Messages {
  132. promptText += fmt.Sprintf("%s: %s\n", message.Role, message.Content)
  133. }
  134. completionText := fmt.Sprintf("%s: %s\n", "assistant", streamResponseText)
  135. quota = countToken(promptText) + countToken(completionText)*completionRatio + 3
  136. } else {
  137. quota = textResponse.Usage.PromptTokens + textResponse.Usage.CompletionTokens*completionRatio
  138. }
  139. ratio := common.GetModelRatio(textRequest.Model)
  140. quota = int(float64(quota) * ratio)
  141. err := model.DecreaseTokenQuota(tokenId, quota)
  142. if err != nil {
  143. common.SysError("Error consuming token remain quota: " + err.Error())
  144. }
  145. }
  146. }()
  147. if isStream {
  148. scanner := bufio.NewScanner(resp.Body)
  149. scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
  150. if atEOF && len(data) == 0 {
  151. return 0, nil, nil
  152. }
  153. if i := strings.Index(string(data), "\n\n"); i >= 0 {
  154. return i + 2, data[0:i], nil
  155. }
  156. if atEOF {
  157. return len(data), data, nil
  158. }
  159. return 0, nil, nil
  160. })
  161. dataChan := make(chan string)
  162. stopChan := make(chan bool)
  163. go func() {
  164. for scanner.Scan() {
  165. data := scanner.Text()
  166. dataChan <- data
  167. data = data[6:]
  168. if data != "[DONE]" {
  169. var streamResponse StreamResponse
  170. err = json.Unmarshal([]byte(data), &streamResponse)
  171. if err != nil {
  172. common.SysError("Error unmarshalling stream response: " + err.Error())
  173. return
  174. }
  175. for _, choice := range streamResponse.Choices {
  176. streamResponseText += choice.Delta.Content
  177. }
  178. }
  179. }
  180. stopChan <- true
  181. }()
  182. c.Writer.Header().Set("Content-Type", "text/event-stream")
  183. c.Writer.Header().Set("Cache-Control", "no-cache")
  184. c.Writer.Header().Set("Connection", "keep-alive")
  185. c.Writer.Header().Set("Transfer-Encoding", "chunked")
  186. c.Stream(func(w io.Writer) bool {
  187. select {
  188. case data := <-dataChan:
  189. c.Render(-1, common.CustomEvent{Data: data})
  190. return true
  191. case <-stopChan:
  192. return false
  193. }
  194. })
  195. err = resp.Body.Close()
  196. if err != nil {
  197. return err
  198. }
  199. return nil
  200. } else {
  201. for k, v := range resp.Header {
  202. c.Writer.Header().Set(k, v[0])
  203. }
  204. if consumeQuota {
  205. responseBody, err := io.ReadAll(resp.Body)
  206. if err != nil {
  207. return err
  208. }
  209. err = resp.Body.Close()
  210. if err != nil {
  211. return err
  212. }
  213. err = json.Unmarshal(responseBody, &textResponse)
  214. if err != nil {
  215. return err
  216. }
  217. // Reset response body
  218. resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
  219. }
  220. _, err = io.Copy(c.Writer, resp.Body)
  221. if err != nil {
  222. return err
  223. }
  224. err = resp.Body.Close()
  225. if err != nil {
  226. return err
  227. }
  228. return nil
  229. }
  230. }
  231. func RelayNotImplemented(c *gin.Context) {
  232. c.JSON(http.StatusOK, gin.H{
  233. "error": gin.H{
  234. "message": "Not Implemented",
  235. "type": "one_api_error",
  236. },
  237. })
  238. }