Merge branch 'main' into latest

Qianxia666 · Nov 15, 2023 · c75a92f · c75a92f
2 parents 15b5db6 + 63cd3f0
commit c75a92f
Show file tree

Hide file tree

Showing 12 changed files with 118 additions and 66 deletions.
diff --git a/.github/workflows/docker-image-amd64.yml b/.github/workflows/docker-image-amd64.yml
@@ -1,54 +1,32 @@
-name: Publish Docker image (amd64)
+name: Docker Image CI
 
 on:
   push:
-    tags:
-      - '*'
-  workflow_dispatch:
-    inputs:
-      name:
-        description: 'reason'
-        required: false
+    branches: [ "main" ]
+  pull_request:
+    branches: [ "main" ]
+
 jobs:
-  push_to_registries:
-    name: Push Docker image to multiple registries
-    runs-on: ubuntu-latest
-    permissions:
-      packages: write
-      contents: read
-    steps:
-      - name: Check out the repo
-        uses: actions/checkout@v3
 
-      - name: Save version info
-        run: |
-          git describe --tags > VERSION 
+  build:
+
+    runs-on: ubuntu-latest
 
-      - name: Log in to Docker Hub
-        uses: docker/login-action@v2
+    steps:
+      - uses: actions/checkout@v3
+      - uses: docker/login-action@v3.0.0
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
-
-      - name: Log in to the Container registry
-        uses: docker/login-action@v2
-        with:
-          registry: ghcr.io
-          username: ${{ github.actor }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-
       - name: Extract metadata (tags, labels) for Docker
         id: meta
-        uses: docker/metadata-action@v4
+        uses: docker/metadata-action@v3
         with:
-          images: |
-            justsong/one-api
-            ghcr.io/${{ github.repository }}
-
-      - name: Build and push Docker images
-        uses: docker/build-push-action@v3
+          images: calciumion/neko-api
+      - name: Build the Docker image
+        uses: docker/build-push-action@v5.0.0
         with:
           context: .
           push: true
           tags: ${{ steps.meta.outputs.tags }}
-          labels: ${{ steps.meta.outputs.labels }}
+          labels: ${{ steps.meta.outputs.labels }}
diff --git a/common/model-ratio.go b/common/model-ratio.go
@@ -37,7 +37,9 @@ var ModelRatio = map[string]float64{
 	"text-davinci-003":          10,
 	"text-davinci-edit-001":     10,
 	"code-davinci-edit-001":     10,
-	"whisper-1":                 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
+	"whisper-1":                 15,  // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
+	"tts-1":                     7.5, // 1k characters -> $0.015
+	"tts-1-hd":                  15,  // 1k characters -> $0.03
 	"davinci":                   10,
 	"curie":                     10,
 	"babbage":                   10,

diff --git a/common/utils.go b/common/utils.go
@@ -207,3 +207,12 @@ func String2Int(str string) int {
 	}
 	return num
 }
+
+func StringsContains(strs []string, str string) bool {
+	for _, s := range strs {
+		if s == str {
+			return true
+		}
+	}
+	return false
+}
diff --git a/controller/relay-audio.go b/controller/relay-audio.go
@@ -11,19 +11,48 @@ import (
 	"net/http"
 	"one-api/common"
 	"one-api/model"
+	"strings"
 )
 
+var availableVoices = []string{
+	"alloy",
+	"echo",
+	"fable",
+	"onyx",
+	"nova",
+	"shimmer",
+}
+
 func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
-	audioModel := "whisper-1"
 
 	tokenId := c.GetInt("token_id")
 	channelType := c.GetInt("channel")
 	channelId := c.GetInt("channel_id")
 	userId := c.GetInt("id")
 	group := c.GetString("group")
 
+	var audioRequest AudioRequest
+	err := common.UnmarshalBodyReusable(c, &audioRequest)
+	if err != nil {
+		return errorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
+	}
+
+	// request validation
+	if audioRequest.Model == "" {
+		return errorWrapper(errors.New("model is required"), "required_field_missing", http.StatusBadRequest)
+	}
+
+	if strings.HasPrefix(audioRequest.Model, "tts-1") {
+		if audioRequest.Voice == "" {
+			return errorWrapper(errors.New("voice is required"), "required_field_missing", http.StatusBadRequest)
+		}
+		if !common.StringsContains(availableVoices, audioRequest.Voice) {
+			return errorWrapper(errors.New("voice must be one of "+strings.Join(availableVoices, ", ")), "invalid_field_value", http.StatusBadRequest)
+		}
+	}
+
 	preConsumedTokens := common.PreConsumedQuota
-	modelRatio := common.GetModelRatio(audioModel)
+	modelRatio := common.GetModelRatio(audioRequest.Model)
 	groupRatio := common.GetGroupRatio(group)
 	ratio := modelRatio * groupRatio
 	preConsumedQuota := int(float64(preConsumedTokens) * ratio)
@@ -58,8 +87,8 @@ func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 		if err != nil {
 			return errorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
 		}
-		if modelMap[audioModel] != "" {
-			audioModel = modelMap[audioModel]
+		if modelMap[audioRequest.Model] != "" {
+			audioRequest.Model = modelMap[audioRequest.Model]
 		}
 	}
 
@@ -97,9 +126,14 @@ func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 
 	defer func(ctx context.Context) {
 		go func() {
-			quota := countTokenText(audioResponse.Text, audioModel)
+			var quota int
+			if strings.HasPrefix(audioRequest.Model, "tts-1") {
+				quota = countAudioToken(audioRequest.Input, audioRequest.Model)
+			} else {
+				quota = countAudioToken(audioResponse.Text, audioRequest.Model)
+			}
 			quotaDelta := quota - preConsumedQuota
-			err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota)
+			err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota, true)
 			if err != nil {
 				common.SysError("error consuming token remain quota: " + err.Error())
 			}
@@ -110,7 +144,7 @@ func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 			if quota != 0 {
 				tokenName := c.GetString("token_name")
 				logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
-				model.RecordConsumeLog(ctx, userId, channelId, 0, 0, audioModel, tokenName, quota, logContent, tokenId)
+				model.RecordConsumeLog(ctx, userId, channelId, 0, 0, audioRequest.Model, tokenName, quota, logContent, tokenId)
 				model.UpdateUserUsedQuotaAndRequestCount(userId, quota)
 				channelId := c.GetInt("channel_id")
 				model.UpdateChannelUsedQuota(channelId, quota)
@@ -127,9 +161,13 @@ func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 	if err != nil {
 		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
 	}
-	err = json.Unmarshal(responseBody, &audioResponse)
-	if err != nil {
-		return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+	if strings.HasPrefix(audioRequest.Model, "tts-1") {
+
+	} else {
+		err = json.Unmarshal(responseBody, &audioResponse)
+		if err != nil {
+			return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+		}
 	}
 
 	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))

diff --git a/controller/relay-image.go b/controller/relay-image.go
@@ -147,7 +147,7 @@ func relayImageHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode
 	var textResponse ImageResponse
 	defer func(ctx context.Context) {
 		if consumeQuota {
-			err := model.PostConsumeTokenQuota(tokenId, userId, quota, 0)
+			err := model.PostConsumeTokenQuota(tokenId, userId, quota, 0, true)
 			if err != nil {
 				common.SysError("error consuming token remain quota: " + err.Error())
 			}

diff --git a/controller/relay-mj.go b/controller/relay-mj.go
@@ -359,7 +359,7 @@ func relayMidjourneySubmit(c *gin.Context, relayMode int) *MidjourneyResponse {
 
 	defer func(ctx context.Context) {
 		if consumeQuota {
-			err := model.PostConsumeTokenQuota(tokenId, userQuota, quota, 0)
+			err := model.PostConsumeTokenQuota(tokenId, userQuota, quota, 0, true)
 			if err != nil {
 				common.SysError("error consuming token remain quota: " + err.Error())
 			}

diff --git a/controller/relay-text.go b/controller/relay-text.go
@@ -400,7 +400,7 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			if preConsumedQuota != 0 {
 				go func(ctx context.Context) {
 					// return pre-consumed quota
-					err := model.PostConsumeTokenQuota(tokenId, userQuota, -preConsumedQuota, 0)
+					err := model.PostConsumeTokenQuota(tokenId, userQuota, -preConsumedQuota, 0, false)
 					if err != nil {
 						common.LogError(ctx, "error return pre-consumed quota: "+err.Error())
 					}
@@ -434,7 +434,7 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 					quota = 0
 				}
 				quotaDelta := quota - preConsumedQuota
-				err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota)
+				err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota, true)
 				if err != nil {
 					common.LogError(ctx, "error consuming token remain quota: "+err.Error())
 				}

diff --git a/controller/relay-utils.go b/controller/relay-utils.go
@@ -10,6 +10,7 @@ import (
 	"one-api/common"
 	"strconv"
 	"strings"
+	"unicode/utf8"
 )
 
 var stopFinishReason = "stop"
@@ -106,6 +107,14 @@ func countTokenInput(input any, model string) int {
 	return 0
 }
 
+func countAudioToken(text string, model string) int {
+	if strings.HasPrefix(model, "tts") {
+		return utf8.RuneCountInString(text)
+	} else {
+		return countTokenText(text, model)
+	}
+}
+
 func countTokenText(text string, model string) int {
 	tokenEncoder := getTokenEncoder(model)
 	return getTokenNum(tokenEncoder, text)

diff --git a/controller/relay.go b/controller/relay.go
@@ -70,6 +70,12 @@ func (r GeneralOpenAIRequest) ParseInput() []string {
 	return input
 }
 
+type AudioRequest struct {
+	Model string `json:"model"`
+	Voice string `json:"voice"`
+	Input string `json:"input"`
+}
+
 type ChatRequest struct {
 	Model     string    `json:"model"`
 	Messages  []Message `json:"messages"`

diff --git a/middleware/distributor.go b/middleware/distributor.go
@@ -46,9 +46,8 @@ func Distribute() func(c *gin.Context) {
 				if modelRequest.Model == "" {
 					modelRequest.Model = "midjourney"
 				}
-			} else if !strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
-				err = common.UnmarshalBodyReusable(c, &modelRequest)
 			}
+			err = common.UnmarshalBodyReusable(c, &modelRequest)
 			if err != nil {
 				abortWithMessage(c, http.StatusBadRequest, "无效的请求")
 				return
@@ -70,7 +69,11 @@ func Distribute() func(c *gin.Context) {
 			}
 			if strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
 				if modelRequest.Model == "" {
-					modelRequest.Model = "whisper-1"
+					if strings.HasPrefix(c.Request.URL.Path, "/v1/audio/speech") {
+						modelRequest.Model = "tts-1"
+					} else {
+						modelRequest.Model = "whisper-1"
+					}
 				}
 			}
 			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model)

diff --git a/model/token.go b/model/token.go
@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"gorm.io/gorm"
 	"one-api/common"
+	"strconv"
 	"strings"
 )
 
@@ -194,22 +195,31 @@ func PreConsumeTokenQuota(tokenId int, quota int) (userQuota int, err error) {
 		return 0, err
 	}
 	if userQuota < quota {
-		return userQuota, errors.New(fmt.Sprintf("用户额度不足，剩余额度为 %d", userQuota))
+		return 0, errors.New(fmt.Sprintf("用户额度不足，剩余额度为 %d", userQuota))
 	}
 	if !token.UnlimitedQuota {
 		err = DecreaseTokenQuota(tokenId, quota)
 		if err != nil {
-			return userQuota, err
+			return 0, err
 		}
 	}
 	err = DecreaseUserQuota(token.UserId, quota)
-	return userQuota, err
+	return userQuota - quota, err
 }
 
-func PostConsumeTokenQuota(tokenId int, userQuota int, quota int, preConsumedQuota int) (err error) {
+func PostConsumeTokenQuota(tokenId int, userQuota int, quota int, preConsumedQuota int, sendEmail bool) (err error) {
 	token, err := GetTokenById(tokenId)
 
 	if quota > 0 {
+		err = DecreaseUserQuota(token.UserId, quota)
+	} else {
+		err = IncreaseUserQuota(token.UserId, -quota)
+	}
+	if err != nil {
+		return err
+	}
+
+	if sendEmail {
 		quotaTooLow := userQuota >= common.QuotaRemindThreshold && userQuota-(quota+preConsumedQuota) < common.QuotaRemindThreshold
 		noMoreQuota := userQuota-(quota+preConsumedQuota) <= 0
 		if quotaTooLow || noMoreQuota {
@@ -229,16 +239,12 @@ func PostConsumeTokenQuota(tokenId int, userQuota int, quota int, preConsumedQuo
 					if err != nil {
 						common.SysError("failed to send email" + err.Error())
 					}
+					common.SysLog("user quota is low, consumed quota: " + strconv.Itoa(quota) + ", user quota: " + strconv.Itoa(userQuota))
 				}
 			}()
 		}
-		err = DecreaseUserQuota(token.UserId, quota)
-	} else {
-		err = IncreaseUserQuota(token.UserId, -quota)
-	}
-	if err != nil {
-		return err
 	}
+
 	if !token.UnlimitedQuota {
 		if quota > 0 {
 			err = DecreaseTokenQuota(tokenId, quota)

diff --git a/router/relay-router.go b/router/relay-router.go
@@ -29,6 +29,7 @@ func SetRelayRouter(router *gin.Engine) {
 		relayV1Router.POST("/engines/:model/embeddings", controller.Relay)
 		relayV1Router.POST("/audio/transcriptions", controller.Relay)
 		relayV1Router.POST("/audio/translations", controller.Relay)
+		relayV1Router.POST("/audio/speech", controller.Relay)
 		relayV1Router.GET("/files", controller.RelayNotImplemented)
 		relayV1Router.POST("/files", controller.RelayNotImplemented)
 		relayV1Router.DELETE("/files/:id", controller.RelayNotImplemented)