feat: add Proxy channel type and relay mode

Add the Proxy channel type and relay mode to support proxying requests to custom upstream services.

.github/workflows/docker-image-amd64-en.yml

@@ -1,4 +1,4 @@
-name: Publish Docker image (English)
+name: Publish Docker image (amd64, English)
 
 on:
   push:
@@ -34,13 +34,6 @@ jobs:
       - name: Translate
         run: |
           python ./i18n/translate.py --repository_path . --json_file_path ./i18n/en.json
-
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v2
-
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-
       - name: Log in to Docker Hub
         uses: docker/login-action@v2
         with:
@@ -58,7 +51,6 @@ jobs:
         uses: docker/build-push-action@v3
         with:
           context: .
-          platforms: linux/amd64,linux/arm64
           push: true
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}

.github/workflows/docker-image-amd64.yml

@@ -0,0 +1,61 @@
+name: Publish Docker image (amd64)
+
+on:
+  push:
+    tags:
+      - 'v*.*.*'
+  workflow_dispatch:
+    inputs:
+      name:
+        description: 'reason'
+        required: false
+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: Check repository URL
+        run: |
+          REPO_URL=$(git config --get remote.origin.url)
+          if [[ $REPO_URL == *"pro" ]]; then
+            exit 1
+          fi        
+
+      - name: Save version info
+        run: |
+          git describe --tags > VERSION 
+
+      - name: Log in to Docker Hub
+        uses: docker/login-action@v2
+        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: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata (tags, labels) for Docker
+        id: meta
+        uses: docker/metadata-action@v4
+        with:
+          images: |
+            justsong/one-api
+            ghcr.io/${{ github.repository }}
+
+      - name: Build and push Docker images
+        uses: docker/build-push-action@v3
+        with:
+          context: .
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}

.github/workflows/docker-image-arm64.yml

@@ -1,9 +1,10 @@
-name: Publish Docker image
+name: Publish Docker image (arm64)
 
 on:
   push:
     tags:
       - 'v*.*.*'
+      - '!*-alpha*'
   workflow_dispatch:
     inputs:
       name:

Dockerfile

@@ -1,4 +1,4 @@
-FROM --platform=$BUILDPLATFORM node:16 AS builder
+FROM node:16 as builder
 
 WORKDIR /web
 COPY ./VERSION .

relay/adaptor/doubao/main.go

@@ -7,12 +7,8 @@ import (
 )
 
 func GetRequestURL(meta *meta.Meta) (string, error) {
-	switch meta.Mode {
-	case relaymode.ChatCompletions:
+	if meta.Mode == relaymode.ChatCompletions {
 		return fmt.Sprintf("%s/api/v3/chat/completions", meta.BaseURL), nil
-	case relaymode.Embeddings:
-		return fmt.Sprintf("%s/api/v3/embeddings", meta.BaseURL), nil
-	default:
 	}
 	return "", fmt.Errorf("unsupported relay mode %d for doubao", meta.Mode)
 }

relay/adaptor/openai/constants.go

@@ -8,7 +8,6 @@ var ModelList = []string{
 	"gpt-4-32k", "gpt-4-32k-0314", "gpt-4-32k-0613",
 	"gpt-4-turbo-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09",
 	"gpt-4o", "gpt-4o-2024-05-13",
-	"gpt-4o-mini", "gpt-4o-mini-2024-07-18",
 	"gpt-4-vision-preview",
 	"text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large",
 	"text-curie-001", "text-babbage-001", "text-ada-001", "text-davinci-002", "text-davinci-003",

relay/adaptor/openai/token.go

@@ -110,7 +110,7 @@ func CountTokenMessages(messages []model.Message, model string) int {
 						if imageUrl["detail"] != nil {
 							detail = imageUrl["detail"].(string)
 						}
-						imageTokens, err := countImageTokens(url, detail, model)
+						imageTokens, err := countImageTokens(url, detail)
 						if err != nil {
 							logger.SysError("error counting image tokens: " + err.Error())
 						} else {
@@ -134,15 +134,11 @@ const (
 	lowDetailCost         = 85
 	highDetailCostPerTile = 170
 	additionalCost        = 85
-	// gpt-4o-mini cost higher than other model
-	gpt4oMiniLowDetailCost  = 2833
-	gpt4oMiniHighDetailCost = 5667
-	gpt4oMiniAdditionalCost = 2833
 )
 
 // https://platform.openai.com/docs/guides/vision/calculating-costs
 // https://github.com/openai/openai-cookbook/blob/05e3f9be4c7a2ae7ecf029a7c32065b024730ebe/examples/How_to_count_tokens_with_tiktoken.ipynb
-func countImageTokens(url string, detail string, model string) (_ int, err error) {
+func countImageTokens(url string, detail string) (_ int, err error) {
 	var fetchSize = true
 	var width, height int
 	// Reference: https://platform.openai.com/docs/guides/vision/low-or-high-fidelity-image-understanding
@@ -176,9 +172,6 @@ func countImageTokens(url string, detail string, model string) (_ int, err error
 	}
 	switch detail {
 	case "low":
-		if strings.HasPrefix(model, "gpt-4o-mini") {
-			return gpt4oMiniLowDetailCost, nil
-		}
 		return lowDetailCost, nil
 	case "high":
 		if fetchSize {
@@ -198,9 +191,6 @@ func countImageTokens(url string, detail string, model string) (_ int, err error
 			height = int(float64(height) * ratio)
 		}
 		numSquares := int(math.Ceil(float64(width)/512) * math.Ceil(float64(height)/512))
-		if strings.HasPrefix(model, "gpt-4o-mini") {
-			return numSquares*gpt4oMiniHighDetailCost + gpt4oMiniAdditionalCost, nil
-		}
 		result := numSquares*highDetailCostPerTile + additionalCost
 		return result, nil
 	default:

relay/billing/ratio/model.go

@@ -28,17 +28,15 @@ var ModelRatio = map[string]float64{
 	"gpt-4-32k":               30,
 	"gpt-4-32k-0314":          30,
 	"gpt-4-32k-0613":          30,
-	"gpt-4-1106-preview":      5,     // $0.01 / 1K tokens
-	"gpt-4-0125-preview":      5,     // $0.01 / 1K tokens
-	"gpt-4-turbo-preview":     5,     // $0.01 / 1K tokens
-	"gpt-4-turbo":             5,     // $0.01 / 1K tokens
-	"gpt-4-turbo-2024-04-09":  5,     // $0.01 / 1K tokens
-	"gpt-4o":                  2.5,   // $0.005 / 1K tokens
-	"gpt-4o-2024-05-13":       2.5,   // $0.005 / 1K tokens
-	"gpt-4o-mini":             0.075, // $0.00015 / 1K tokens
-	"gpt-4o-mini-2024-07-18":  0.075, // $0.00015 / 1K tokens
-	"gpt-4-vision-preview":    5,     // $0.01 / 1K tokens
-	"gpt-3.5-turbo":           0.25,  // $0.0005 / 1K tokens
+	"gpt-4-1106-preview":      5,    // $0.01 / 1K tokens
+	"gpt-4-0125-preview":      5,    // $0.01 / 1K tokens
+	"gpt-4-turbo-preview":     5,    // $0.01 / 1K tokens
+	"gpt-4-turbo":             5,    // $0.01 / 1K tokens
+	"gpt-4-turbo-2024-04-09":  5,    // $0.01 / 1K tokens
+	"gpt-4o":                  2.5,  // $0.005 / 1K tokens
+	"gpt-4o-2024-05-13":       2.5,  // $0.005 / 1K tokens
+	"gpt-4-vision-preview":    5,    // $0.01 / 1K tokens
+	"gpt-3.5-turbo":           0.25, // $0.0005 / 1K tokens
 	"gpt-3.5-turbo-0301":      0.75,
 	"gpt-3.5-turbo-0613":      0.75,
 	"gpt-3.5-turbo-16k":       1.5, // $0.003 / 1K tokens
@@ -310,9 +308,6 @@ func GetCompletionRatio(name string, channelType int) float64 {
 		return 4.0 / 3.0
 	}
 	if strings.HasPrefix(name, "gpt-4") {
-		if strings.HasPrefix(name, "gpt-4o-mini") {
-			return 4
-		}
 		if strings.HasPrefix(name, "gpt-4-turbo") ||
 			strings.HasPrefix(name, "gpt-4o") ||
 			strings.HasSuffix(name, "preview") {

relay/controller/text.go

@@ -10,7 +10,6 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay"
-	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/apitype"
 	"github.com/songquanpeng/one-api/relay/billing"
@@ -32,8 +31,9 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	meta.IsStream = textRequest.Stream
 
 	// map model name
+	var isModelMapped bool
 	meta.OriginModelName = textRequest.Model
-	textRequest.Model, _ = getMappedModelName(textRequest.Model, meta.ModelMapping)
+	textRequest.Model, isModelMapped = getMappedModelName(textRequest.Model, meta.ModelMapping)
 	meta.ActualModelName = textRequest.Model
 	// get model ratio & group ratio
 	modelRatio := billingratio.GetModelRatio(textRequest.Model, meta.ChannelType)
@@ -55,9 +55,30 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	adaptor.Init(meta)
 
 	// get request body
-	requestBody, err := getRequestBody(c, meta, textRequest, adaptor)
-	if err != nil {
-		return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
+	var requestBody io.Reader
+	if meta.APIType == apitype.OpenAI {
+		// no need to convert request for openai
+		shouldResetRequestBody := isModelMapped || meta.ChannelType == channeltype.Baichuan // frequency_penalty 0 is not acceptable for baichuan
+		if shouldResetRequestBody {
+			jsonStr, err := json.Marshal(textRequest)
+			if err != nil {
+				return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
+			}
+			requestBody = bytes.NewBuffer(jsonStr)
+		} else {
+			requestBody = c.Request.Body
+		}
+	} else {
+		convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
+		if err != nil {
+			return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
+		}
+		jsonData, err := json.Marshal(convertedRequest)
+		if err != nil {
+			return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
+		}
+		logger.Debugf(ctx, "converted request: \n%s", string(jsonData))
+		requestBody = bytes.NewBuffer(jsonData)
 	}
 
 	// do request
@@ -82,26 +103,3 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	go postConsumeQuota(ctx, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio)
 	return nil
 }
-
-func getRequestBody(c *gin.Context, meta *meta.Meta, textRequest *model.GeneralOpenAIRequest, adaptor adaptor.Adaptor) (io.Reader, error) {
-	if meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
-		// no need to convert request for openai
-		return c.Request.Body, nil
-	}
-
-	// get request body
-	var requestBody io.Reader
-	convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
-	if err != nil {
-		logger.Debugf(c.Request.Context(), "converted request failed: %s\n", err.Error())
-		return nil, err
-	}
-	jsonData, err := json.Marshal(convertedRequest)
-	if err != nil {
-		logger.Debugf(c.Request.Context(), "converted request json_marshal_failed: %s\n", err.Error())
-		return nil, err
-	}
-	logger.Debugf(c.Request.Context(), "converted request: \n%s", string(jsonData))
-	requestBody = bytes.NewBuffer(jsonData)
-	return requestBody, nil
-}