fix: change GetAudioTokens to return float64 and update related functions

relay/adaptor/openai/token.go

@@ -93,7 +93,9 @@ func CountTokenMessages(ctx context.Context,
 		tokensPerMessage = 3
 		tokensPerName = 1
 	}
+
 	tokenNum := 0
+	var totalAudioTokens float64
 	for _, message := range messages {
 		tokenNum += tokensPerMessage
 		contents := message.ParseContent()
@@ -117,17 +119,19 @@ func CountTokenMessages(ctx context.Context,
 					logger.SysError("error decoding audio data: " + err.Error())
 				}
 
-				tokens, err := helper.GetAudioTokens(ctx,
+				audioTokens, err := helper.GetAudioTokens(ctx,
 					bytes.NewReader(audioData),
 					ratio.GetAudioPromptTokensPerSecond(actualModel))
 				if err != nil {
 					logger.SysError("error counting audio tokens: " + err.Error())
 				} else {
-					tokenNum += tokens
+					audioTokens += audioTokens
 				}
 			}
 		}
 
+		tokenNum += int(math.Ceil(totalAudioTokens))
+
 		tokenNum += getTokenNum(tokenEncoder, message.Role)
 		if message.Name != nil {
 			tokenNum += tokensPerName

relay/controller/audio.go

@@ -7,6 +7,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"math"
 	"mime/multipart"
 	"net/http"
 	"strings"
@@ -33,7 +34,7 @@ type commonAudioRequest struct {
 	File *multipart.FileHeader `form:"file" binding:"required"`
 }
 
-func countAudioTokens(c *gin.Context) (int, error) {
+func countAudioTokens(c *gin.Context) (float64, error) {
 	body, err := common.GetRequestBody(c)
 	if err != nil {
 		return 0, errors.WithStack(err)
@@ -102,7 +103,7 @@ func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 			return openai.ErrorWrapper(err, "count_audio_tokens_failed", http.StatusInternalServerError)
 		}
 
-		preConsumedQuota = int64(float64(audioTokens) * ratio)
+		preConsumedQuota = int64(math.Ceil(audioTokens * ratio))
 		quota = preConsumedQuota
 	default:
 		return openai.ErrorWrapper(errors.New("unexpected_relay_mode"), "unexpected_relay_mode", http.StatusInternalServerError)