feat: support claude-3 (close #1080, close #1094)

百度千帆平台上的第三方大模型调用

.github/workflows/linux-release.yml

@@ -23,7 +23,7 @@ jobs:
       - uses: actions/setup-node@v3
         with:
           node-version: 16
-      - name: Build Frontend (theme default)
+      - name: Build Frontend
         env:
           CI: ""
         run: |
@@ -38,7 +38,7 @@ jobs:
       - name: Build Backend (amd64)
         run: |
           go mod download
-          go build -ldflags "-s -w -X 'one-api/common.Version=$(git describe --tags)' -extldflags '-static'" -o one-api
+          go build -ldflags "-s -w -X 'github.com/songquanpeng/one-api/common.Version=$(git describe --tags)' -extldflags '-static'" -o one-api
 
       - name: Build Backend (arm64)
         run: |

.github/workflows/macos-release.yml

@@ -23,7 +23,7 @@ jobs:
       - uses: actions/setup-node@v3
         with:
           node-version: 16
-      - name: Build Frontend (theme default)
+      - name: Build Frontend
         env:
           CI: ""
         run: |
@@ -38,7 +38,7 @@ jobs:
       - name: Build Backend
         run: |
           go mod download
-          go build -ldflags "-X 'one-api/common.Version=$(git describe --tags)'" -o one-api-macos
+          go build -ldflags "-X 'github.com/songquanpeng/one-api/common.Version=$(git describe --tags)'" -o one-api-macos
       - name: Release
         uses: softprops/action-gh-release@v1
         if: startsWith(github.ref, 'refs/tags/')

.github/workflows/windows-release.yml

@@ -26,7 +26,7 @@ jobs:
       - uses: actions/setup-node@v3
         with:
           node-version: 16
-      - name: Build Frontend (theme default)
+      - name: Build Frontend
         env:
           CI: ""
         run: |
@@ -41,7 +41,7 @@ jobs:
       - name: Build Backend
         run: |
           go mod download
-          go build -ldflags "-s -w -X 'one-api/common.Version=$(git describe --tags)'" -o one-api.exe
+          go build -ldflags "-s -w -X 'github.com/songquanpeng/one-api/common.Version=$(git describe --tags)'" -o one-api.exe
       - name: Release
         uses: softprops/action-gh-release@v1
         if: startsWith(github.ref, 'refs/tags/')

Dockerfile

@@ -23,7 +23,7 @@ ADD go.mod go.sum ./
 RUN go mod download
 COPY . .
 COPY --from=builder /web/build ./web/build
-RUN go build -ldflags "-s -w -X 'one-api/common.Version=$(cat VERSION)' -extldflags '-static'" -o one-api
+RUN go build -ldflags "-s -w -X 'github.com/songquanpeng/one-api/common.Version=$(cat VERSION)' -extldflags '-static'" -o one-api
 
 FROM alpine
 

README.md

@@ -67,6 +67,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [OpenAI ChatGPT 系列模型](https://platform.openai.com/docs/guides/gpt/chat-completions-api)（支持 [Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference)）
    + [x] [Anthropic Claude 系列模型](https://anthropic.com)
    + [x] [Google PaLM2/Gemini 系列模型](https://developers.generativeai.google)
+   + [x] [Mistral 系列模型](https://mistral.ai/)
    + [x] [百度文心一言系列模型](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
    + [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
    + [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)
@@ -74,8 +75,9 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [360 智脑](https://ai.360.cn)
    + [x] [腾讯混元大模型](https://cloud.tencent.com/document/product/1729)
    + [x] [Moonshot AI](https://platform.moonshot.cn/)
+   + [x] [百川大模型](https://platform.baichuan-ai.com)
    + [ ] [字节云雀大模型](https://www.volcengine.com/product/ark) (WIP)
-   + [ ] [MINIMAX](https://api.minimax.chat/) (WIP)
+   + [x] [MINIMAX](https://api.minimax.chat/)
 2. 支持配置镜像以及众多[第三方代理服务](https://iamazing.cn/page/openai-api-third-party-services)。
 3. 支持通过**负载均衡**的方式访问多个渠道。
 4. 支持 **stream 模式**，可以通过流式传输实现打字机效果。

common/constants.go

@@ -64,6 +64,9 @@ const (
 	ChannelTypeTencent        = 23
 	ChannelTypeGemini         = 24
 	ChannelTypeMoonshot       = 25
+	ChannelTypeBaichuan       = 26
+	ChannelTypeMinimax        = 27
+	ChannelTypeMistral        = 28
 )
 
 var ChannelBaseURLs = []string{
@@ -93,6 +96,9 @@ var ChannelBaseURLs = []string{
 	"https://hunyuan.cloud.tencent.com",         // 23
 	"https://generativelanguage.googleapis.com", // 24
 	"https://api.moonshot.cn",                   // 25
+	"https://api.baichuan-ai.com",               // 26
+	"https://api.minimax.chat",                  // 27
+	"https://api.mistral.ai",                    // 28
 }
 
 const (

common/gin.go

@@ -8,12 +8,24 @@ import (
 	"strings"
 )
 
-func UnmarshalBodyReusable(c *gin.Context, v any) error {
+const KeyRequestBody = "key_request_body"
+
+func GetRequestBody(c *gin.Context) ([]byte, error) {
+	requestBody, _ := c.Get(KeyRequestBody)
+	if requestBody != nil {
+		return requestBody.([]byte), nil
+	}
 	requestBody, err := io.ReadAll(c.Request.Body)
 	if err != nil {
-		return err
+		return nil, err
 	}
-	err = c.Request.Body.Close()
+	_ = c.Request.Body.Close()
+	c.Set(KeyRequestBody, requestBody)
+	return requestBody.([]byte), nil
+}
+
+func UnmarshalBodyReusable(c *gin.Context, v any) error {
+	requestBody, err := GetRequestBody(c)
 	if err != nil {
 		return err
 	}

common/helper/helper.go

@@ -137,6 +137,7 @@ func GetUUID() string {
 }
 
 const keyChars = "0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
+const keyNumbers = "0123456789"
 
 func init() {
 	rand.Seed(time.Now().UnixNano())
@@ -168,6 +169,15 @@ func GetRandomString(length int) string {
 	return string(key)
 }
 
+func GetRandomNumberString(length int) string {
+	rand.Seed(time.Now().UnixNano())
+	key := make([]byte, length)
+	for i := 0; i < length; i++ {
+		key[i] = keyNumbers[rand.Intn(len(keyNumbers))]
+	}
+	return string(key)
+}
+
 func GetTimestamp() int64 {
 	return time.Now().Unix()
 }

common/logger/logger.go

@@ -13,6 +13,7 @@ import (
 )
 
 const (
+	loggerDEBUG = "DEBUG"
 	loggerINFO  = "INFO"
 	loggerWarn  = "WARN"
 	loggerError = "ERR"
@@ -55,6 +56,10 @@ func SysError(s string) {
 	_, _ = fmt.Fprintf(gin.DefaultErrorWriter, "[SYS] %v | %s \n", t.Format("2006/01/02 - 15:04:05"), s)
 }
 
+func Debug(ctx context.Context, msg string) {
+	logHelper(ctx, loggerDEBUG, msg)
+}
+
 func Info(ctx context.Context, msg string) {
 	logHelper(ctx, loggerINFO, msg)
 }
@@ -67,6 +72,10 @@ func Error(ctx context.Context, msg string) {
 	logHelper(ctx, loggerError, msg)
 }
 
+func Debugf(ctx context.Context, format string, a ...any) {
+	Debug(ctx, fmt.Sprintf(format, a...))
+}
+
 func Infof(ctx context.Context, format string, a ...any) {
 	Info(ctx, fmt.Sprintf(format, a...))
 }

common/model-ratio.go

@@ -7,29 +7,6 @@ import (
 	"time"
 )
 
-var DalleSizeRatios = map[string]map[string]float64{
-	"dall-e-2": {
-		"256x256":   1,
-		"512x512":   1.125,
-		"1024x1024": 1.25,
-	},
-	"dall-e-3": {
-		"1024x1024": 1,
-		"1024x1792": 2,
-		"1792x1024": 2,
-	},
-}
-
-var DalleGenerationImageAmounts = map[string][2]int{
-	"dall-e-2": {1, 10},
-	"dall-e-3": {1, 1}, // OpenAI allows n=1 currently.
-}
-
-var DalleImagePromptLengthLimitations = map[string]int{
-	"dall-e-2": 1000,
-	"dall-e-3": 4000,
-}
-
 const (
 	USD2RMB = 7
 	USD     = 500 // $0.002 = 1 -> $1 = 500
@@ -40,7 +17,6 @@ const (
 // https://platform.openai.com/docs/models/model-endpoint-compatibility
 // https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
 // https://openai.com/pricing
-// TODO: when a new api is enabled, check the pricing here
 // 1 === $0.002 / 1K tokens
 // 1 === ￥0.014 / 1k tokens
 var ModelRatio = map[string]float64{
@@ -87,21 +63,28 @@ var ModelRatio = map[string]float64{
 	"text-search-ada-doc-001": 10,
 	"text-moderation-stable":  0.1,
 	"text-moderation-latest":  0.1,
-	"dall-e-2":                8,     // $0.016 - $0.020 / image
-	"dall-e-3":                20,    // $0.040 - $0.120 / image
-	"claude-instant-1":        0.815, // $1.63 / 1M tokens
-	"claude-2":                5.51,  // $11.02 / 1M tokens
-	"claude-2.0":              5.51,  // $11.02 / 1M tokens
-	"claude-2.1":              5.51,  // $11.02 / 1M tokens
+	"dall-e-2":                8,  // $0.016 - $0.020 / image
+	"dall-e-3":                20, // $0.040 - $0.120 / image
+	// https://www.anthropic.com/api#pricing
+	"claude-instant-1.2":       0.8 / 1000 * USD,
+	"claude-2.0":               8.0 / 1000 * USD,
+	"claude-2.1":               8.0 / 1000 * USD,
+	"claude-3-haiku-20240229":  0.25 / 1000 * USD,
+	"claude-3-sonnet-20240229": 3.0 / 1000 * USD,
+	"claude-3-opus-20240229":   15.0 / 1000 * USD,
 	// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7
-	"ERNIE-Bot":                 0.8572,     // ￥0.012 / 1k tokens
-	"ERNIE-Bot-turbo":           0.5715,     // ￥0.008 / 1k tokens
-	"ERNIE-Bot-4":               0.12 * RMB, // ￥0.12 / 1k tokens
-	"ERNIE-Bot-8k":              0.024 * RMB,
-	"Embedding-V1":              0.1429, // ￥0.002 / 1k tokens
-	"PaLM-2":                    1,
-	"gemini-pro":                1,      // $0.00025 / 1k characters -> $0.001 / 1k tokens
-	"gemini-pro-vision":         1,      // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	"ERNIE-Bot":         0.8572,     // ￥0.012 / 1k tokens
+	"ERNIE-Bot-turbo":   0.5715,     // ￥0.008 / 1k tokens
+	"ERNIE-Bot-4":       0.12 * RMB, // ￥0.12 / 1k tokens
+	"ERNIE-Bot-8k":      0.024 * RMB,
+	"Embedding-V1":      0.1429, // ￥0.002 / 1k tokens
+	"PaLM-2":            1,
+	"gemini-pro":        1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	"gemini-pro-vision": 1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	// https://open.bigmodel.cn/pricing
+	"glm-4":                     0.1 * RMB,
+	"glm-4v":                    0.1 * RMB,
+	"glm-3-turbo":               0.005 * RMB,
 	"chatglm_turbo":             0.3572, // ￥0.005 / 1k tokens
 	"chatglm_pro":               0.7143, // ￥0.01 / 1k tokens
 	"chatglm_std":               0.3572, // ￥0.005 / 1k tokens
@@ -127,6 +110,37 @@ var ModelRatio = map[string]float64{
 	"moonshot-v1-8k":   0.012 * RMB,
 	"moonshot-v1-32k":  0.024 * RMB,
 	"moonshot-v1-128k": 0.06 * RMB,
+	// https://platform.baichuan-ai.com/price
+	"Baichuan2-Turbo":      0.008 * RMB,
+	"Baichuan2-Turbo-192k": 0.016 * RMB,
+	"Baichuan2-53B":        0.02 * RMB,
+	// https://api.minimax.chat/document/price
+	"abab6-chat":    0.1 * RMB,
+	"abab5.5-chat":  0.015 * RMB,
+	"abab5.5s-chat": 0.005 * RMB,
+	// https://docs.mistral.ai/platform/pricing/
+	"open-mistral-7b":       0.25 / 1000 * USD,
+	"open-mixtral-8x7b":     0.7 / 1000 * USD,
+	"mistral-small-latest":  2.0 / 1000 * USD,
+	"mistral-medium-latest": 2.7 / 1000 * USD,
+	"mistral-large-latest":  8.0 / 1000 * USD,
+	"mistral-embed":         0.1 / 1000 * USD,
+}
+
+var CompletionRatio = map[string]float64{}
+
+var DefaultModelRatio map[string]float64
+var DefaultCompletionRatio map[string]float64
+
+func init() {
+	DefaultModelRatio = make(map[string]float64)
+	for k, v := range ModelRatio {
+		DefaultModelRatio[k] = v
+	}
+	DefaultCompletionRatio = make(map[string]float64)
+	for k, v := range CompletionRatio {
+		DefaultCompletionRatio[k] = v
+	}
 }
 
 func ModelRatio2JSONString() string {
@@ -147,6 +161,9 @@ func GetModelRatio(name string) float64 {
 		name = strings.TrimSuffix(name, "-internet")
 	}
 	ratio, ok := ModelRatio[name]
+	if !ok {
+		ratio, ok = DefaultModelRatio[name]
+	}
 	if !ok {
 		logger.SysError("model ratio not found: " + name)
 		return 30
@@ -154,8 +171,6 @@ func GetModelRatio(name string) float64 {
 	return ratio
 }
 
-var CompletionRatio = map[string]float64{}
-
 func CompletionRatio2JSONString() string {
 	jsonBytes, err := json.Marshal(CompletionRatio)
 	if err != nil {
@@ -173,6 +188,9 @@ func GetCompletionRatio(name string) float64 {
 	if ratio, ok := CompletionRatio[name]; ok {
 		return ratio
 	}
+	if ratio, ok := DefaultCompletionRatio[name]; ok {
+		return ratio
+	}
 	if strings.HasPrefix(name, "gpt-3.5") {
 		if strings.HasSuffix(name, "0125") {
 			// https://openai.com/blog/new-embedding-models-and-api-updates
@@ -199,11 +217,14 @@ func GetCompletionRatio(name string) float64 {
 		}
 		return 2
 	}
-	if strings.HasPrefix(name, "claude-instant-1") {
-		return 3.38
+	if strings.HasPrefix(name, "claude-3") {
+		return 5
 	}
-	if strings.HasPrefix(name, "claude-2") {
-		return 2.965517
+	if strings.HasPrefix(name, "claude-") {
+		return 3
+	}
+	if strings.HasPrefix(name, "mistral-") {
+		return 3
 	}
 	return 1
 }

common/random.go

@@ -0,0 +1,8 @@
+package common
+
+import "math/rand"
+
+// RandRange returns a random number between min and max (max is not included)
+func RandRange(min, max int) int {
+	return min + rand.Intn(max-min)
+}

controller/channel-test.go

@@ -8,6 +8,7 @@ import (
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/middleware"
 	"github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/helper"
@@ -18,6 +19,7 @@ import (
 	"net/http/httptest"
 	"net/url"
 	"strconv"
+	"strings"
 	"sync"
 	"time"
 
@@ -51,6 +53,7 @@ func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error
 	c.Request.Header.Set("Content-Type", "application/json")
 	c.Set("channel", channel.Type)
 	c.Set("base_url", channel.GetBaseURL())
+	middleware.SetupContextForSelectedChannel(c, channel, "")
 	meta := util.GetRelayMeta(c)
 	apiType := constant.ChannelType2APIType(channel.Type)
 	adaptor := helper.GetAdaptor(apiType)
@@ -59,6 +62,12 @@ func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error
 	}
 	adaptor.Init(meta)
 	modelName := adaptor.GetModelList()[0]
+	if !strings.Contains(channel.Models, modelName) {
+		modelNames := strings.Split(channel.Models, ",")
+		if len(modelNames) > 0 {
+			modelName = modelNames[0]
+		}
+	}
 	request := buildTestRequest()
 	request.Model = modelName
 	meta.OriginModelName, meta.ActualModelName = modelName, modelName

controller/model.go

@@ -4,6 +4,9 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
+	"github.com/songquanpeng/one-api/relay/channel/baichuan"
+	"github.com/songquanpeng/one-api/relay/channel/minimax"
+	"github.com/songquanpeng/one-api/relay/channel/mistral"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/helper"
@@ -98,6 +101,39 @@ func init() {
 			Parent:     nil,
 		})
 	}
+	for _, modelName := range baichuan.ModelList {
+		openAIModels = append(openAIModels, OpenAIModels{
+			Id:         modelName,
+			Object:     "model",
+			Created:    1626777600,
+			OwnedBy:    "baichuan",
+			Permission: permission,
+			Root:       modelName,
+			Parent:     nil,
+		})
+	}
+	for _, modelName := range minimax.ModelList {
+		openAIModels = append(openAIModels, OpenAIModels{
+			Id:         modelName,
+			Object:     "model",
+			Created:    1626777600,
+			OwnedBy:    "minimax",
+			Permission: permission,
+			Root:       modelName,
+			Parent:     nil,
+		})
+	}
+	for _, modelName := range mistral.ModelList {
+		openAIModels = append(openAIModels, OpenAIModels{
+			Id:         modelName,
+			Object:     "model",
+			Created:    1626777600,
+			OwnedBy:    "mistralai",
+			Permission: permission,
+			Root:       modelName,
+			Parent:     nil,
+		})
+	}
 	openAIModelsMap = make(map[string]OpenAIModels)
 	for _, model := range openAIModels {
 		openAIModelsMap[model.Id] = model

controller/relay.go

@@ -1,23 +1,27 @@
 package controller
 
 import (
+	"bytes"
+	"context"
 	"fmt"
 	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/middleware"
+	dbmodel "github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/controller"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
+	"io"
 	"net/http"
-	"strconv"
 )
 
 // https://platform.openai.com/docs/api-reference/chat
 
-func Relay(c *gin.Context) {
-	relayMode := constant.Path2RelayMode(c.Request.URL.Path)
+func relay(c *gin.Context, relayMode int) *model.ErrorWithStatusCode {
 	var err *model.ErrorWithStatusCode
 	switch relayMode {
 	case constant.RelayModeImagesGenerations:
@@ -31,32 +35,89 @@ func Relay(c *gin.Context) {
 	default:
 		err = controller.RelayTextHelper(c)
 	}
-	if err != nil {
-		requestId := c.GetString(logger.RequestIdKey)
-		retryTimesStr := c.Query("retry")
-		retryTimes, _ := strconv.Atoi(retryTimesStr)
-		if retryTimesStr == "" {
-			retryTimes = config.RetryTimes
+	return err
+}
+
+func Relay(c *gin.Context) {
+	ctx := c.Request.Context()
+	relayMode := constant.Path2RelayMode(c.Request.URL.Path)
+	if config.DebugEnabled {
+		requestBody, _ := common.GetRequestBody(c)
+		logger.Debugf(ctx, "request body: %s", string(requestBody))
+	}
+	bizErr := relay(c, relayMode)
+	if bizErr == nil {
+		return
+	}
+	channelId := c.GetInt("channel_id")
+	lastFailedChannelId := channelId
+	channelName := c.GetString("channel_name")
+	group := c.GetString("group")
+	originalModel := c.GetString("original_model")
+	go processChannelRelayError(ctx, channelId, channelName, bizErr)
+	requestId := c.GetString(logger.RequestIdKey)
+	retryTimes := config.RetryTimes
+	if !shouldRetry(c, bizErr.StatusCode) {
+		logger.Errorf(ctx, "relay error happen, status code is %d, won't retry in this case", bizErr.StatusCode)
+		retryTimes = 0
+	}
+	for i := retryTimes; i > 0; i-- {
+		channel, err := dbmodel.CacheGetRandomSatisfiedChannel(group, originalModel, i != retryTimes)
+		if err != nil {
+			logger.Errorf(ctx, "CacheGetRandomSatisfiedChannel failed: %w", err)
+			break
 		}
-		if retryTimes > 0 {
-			c.Redirect(http.StatusTemporaryRedirect, fmt.Sprintf("%s?retry=%d", c.Request.URL.Path, retryTimes-1))
-		} else {
-			if err.StatusCode == http.StatusTooManyRequests {
-				err.Error.Message = "当前分组上游负载已饱和，请稍后再试"
-			}
-			err.Error.Message = helper.MessageWithRequestId(err.Error.Message, requestId)
-			c.JSON(err.StatusCode, gin.H{
-				"error": err.Error,
-			})
+		logger.Infof(ctx, "using channel #%d to retry (remain times %d)", channel.Id, i)
+		if channel.Id == lastFailedChannelId {
+			continue
+		}
+		middleware.SetupContextForSelectedChannel(c, channel, originalModel)
+		requestBody, err := common.GetRequestBody(c)
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(requestBody))
+		bizErr = relay(c, relayMode)
+		if bizErr == nil {
+			return
 		}
 		channelId := c.GetInt("channel_id")
-		logger.Error(c.Request.Context(), fmt.Sprintf("relay error (channel #%d): %s", channelId, err.Message))
-		// https://platform.openai.com/docs/guides/error-codes/api-errors
-		if util.ShouldDisableChannel(&err.Error, err.StatusCode) {
-			channelId := c.GetInt("channel_id")
-			channelName := c.GetString("channel_name")
-			disableChannel(channelId, channelName, err.Message)
+		lastFailedChannelId = channelId
+		channelName := c.GetString("channel_name")
+		go processChannelRelayError(ctx, channelId, channelName, bizErr)
+	}
+	if bizErr != nil {
+		if bizErr.StatusCode == http.StatusTooManyRequests {
+			bizErr.Error.Message = "当前分组上游负载已饱和，请稍后再试"
 		}
+		bizErr.Error.Message = helper.MessageWithRequestId(bizErr.Error.Message, requestId)
+		c.JSON(bizErr.StatusCode, gin.H{
+			"error": bizErr.Error,
+		})
+	}
+}
+
+func shouldRetry(c *gin.Context, statusCode int) bool {
+	if _, ok := c.Get("specific_channel_id"); ok {
+		return false
+	}
+	if statusCode == http.StatusTooManyRequests {
+		return true
+	}
+	if statusCode/100 == 5 {
+		return true
+	}
+	if statusCode == http.StatusBadRequest {
+		return false
+	}
+	if statusCode/100 == 2 {
+		return false
+	}
+	return true
+}
+
+func processChannelRelayError(ctx context.Context, channelId int, channelName string, err *model.ErrorWithStatusCode) {
+	logger.Errorf(ctx, "relay error (channel #%d): %s", channelId, err.Message)
+	// https://platform.openai.com/docs/guides/error-codes/api-errors
+	if util.ShouldDisableChannel(&err.Error, err.StatusCode) {
+		disableChannel(channelId, channelName, err.Message)
 	}
 }
 

i18n/en.json

@@ -456,6 +456,7 @@
   "已绑定的邮箱账户": "Email Account Bound",
   "用户信息更新成功！": "User information updated successfully!",
   "模型倍率 %.2f，分组倍率 %.2f": "model rate %.2f, group rate %.2f",
+  "模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f": "model rate %.2f, group rate %.2f, completion rate %.2f",
   "使用明细（总消耗额度：{renderQuota(stat.quota)}）": "Usage Details (Total Consumption Quota: {renderQuota(stat.quota)})",
   "用户名称": "User Name",
   "令牌名称": "Token Name",

middleware/auth.go

@@ -108,7 +108,7 @@ func TokenAuth() func(c *gin.Context) {
 		c.Set("token_name", token.Name)
 		if len(parts) > 1 {
 			if model.IsAdmin(token.UserId) {
-				c.Set("channelId", parts[1])
+				c.Set("specific_channel_id", parts[1])
 			} else {
 				abortWithMessage(c, http.StatusForbidden, "普通用户不支持指定渠道")
 				return

middleware/distributor.go

@@ -21,8 +21,9 @@ func Distribute() func(c *gin.Context) {
 		userId := c.GetInt("id")
 		userGroup, _ := model.CacheGetUserGroup(userId)
 		c.Set("group", userGroup)
+		var requestModel string
 		var channel *model.Channel
-		channelId, ok := c.Get("channelId")
+		channelId, ok := c.Get("specific_channel_id")
 		if ok {
 			id, err := strconv.Atoi(channelId.(string))
 			if err != nil {
@@ -66,7 +67,8 @@ func Distribute() func(c *gin.Context) {
 					modelRequest.Model = "whisper-1"
 				}
 			}
-			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model)
+			requestModel = modelRequest.Model
+			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model, false)
 			if err != nil {
 				message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", userGroup, modelRequest.Model)
 				if channel != nil {
@@ -77,29 +79,34 @@ func Distribute() func(c *gin.Context) {
 				return
 			}
 		}
-		c.Set("channel", channel.Type)
-		c.Set("channel_id", channel.Id)
-		c.Set("channel_name", channel.Name)
-		c.Set("model_mapping", channel.GetModelMapping())
-		c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))
-		c.Set("base_url", channel.GetBaseURL())
-		// this is for backward compatibility
-		switch channel.Type {
-		case common.ChannelTypeAzure:
-			c.Set(common.ConfigKeyAPIVersion, channel.Other)
-		case common.ChannelTypeXunfei:
-			c.Set(common.ConfigKeyAPIVersion, channel.Other)
-		case common.ChannelTypeGemini:
-			c.Set(common.ConfigKeyAPIVersion, channel.Other)
-		case common.ChannelTypeAIProxyLibrary:
-			c.Set(common.ConfigKeyLibraryID, channel.Other)
-		case common.ChannelTypeAli:
-			c.Set(common.ConfigKeyPlugin, channel.Other)
-		}
-		cfg, _ := channel.LoadConfig()
-		for k, v := range cfg {
-			c.Set(common.ConfigKeyPrefix+k, v)
-		}
+		SetupContextForSelectedChannel(c, channel, requestModel)
 		c.Next()
 	}
 }
+
+func SetupContextForSelectedChannel(c *gin.Context, channel *model.Channel, modelName string) {
+	c.Set("channel", channel.Type)
+	c.Set("channel_id", channel.Id)
+	c.Set("channel_name", channel.Name)
+	c.Set("model_mapping", channel.GetModelMapping())
+	c.Set("original_model", modelName) // for retry
+	c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))
+	c.Set("base_url", channel.GetBaseURL())
+	// this is for backward compatibility
+	switch channel.Type {
+	case common.ChannelTypeAzure:
+		c.Set(common.ConfigKeyAPIVersion, channel.Other)
+	case common.ChannelTypeXunfei:
+		c.Set(common.ConfigKeyAPIVersion, channel.Other)
+	case common.ChannelTypeGemini:
+		c.Set(common.ConfigKeyAPIVersion, channel.Other)
+	case common.ChannelTypeAIProxyLibrary:
+		c.Set(common.ConfigKeyLibraryID, channel.Other)
+	case common.ChannelTypeAli:
+		c.Set(common.ConfigKeyPlugin, channel.Other)
+	}
+	cfg, _ := channel.LoadConfig()
+	for k, v := range cfg {
+		c.Set(common.ConfigKeyPrefix+k, v)
+	}
+}

middleware/request-id.go

@@ -9,7 +9,7 @@ import (
 
 func RequestId() func(c *gin.Context) {
 	return func(c *gin.Context) {
-		id := helper.GetTimeString() + helper.GetRandomString(8)
+		id := helper.GetTimeString() + helper.GetRandomNumberString(8)
 		c.Set(logger.RequestIdKey, id)
 		ctx := context.WithValue(c.Request.Context(), logger.RequestIdKey, id)
 		c.Request = c.Request.WithContext(ctx)

model/cache.go

@@ -94,7 +94,7 @@ func CacheUpdateUserQuota(id int) error {
 	if !common.RedisEnabled {
 		return nil
 	}
-	quota, err := GetUserQuota(id)
+	quota, err := CacheGetUserQuota(id)
 	if err != nil {
 		return err
 	}
@@ -191,7 +191,7 @@ func SyncChannelCache(frequency int) {
 	}
 }
 
-func CacheGetRandomSatisfiedChannel(group string, model string) (*Channel, error) {
+func CacheGetRandomSatisfiedChannel(group string, model string, ignoreFirstPriority bool) (*Channel, error) {
 	if !config.MemoryCacheEnabled {
 		return GetRandomSatisfiedChannel(group, model)
 	}
@@ -213,5 +213,10 @@ func CacheGetRandomSatisfiedChannel(group string, model string) (*Channel, error
 		}
 	}
 	idx := rand.Intn(endIdx)
+	if ignoreFirstPriority {
+		if endIdx < len(channels) { // which means there are more than one priority
+			idx = common.RandRange(endIdx, len(channels))
+		}
+	}
 	return channels[idx], nil
 }

relay/channel/aiproxy/main.go

@@ -53,7 +53,7 @@ func responseAIProxyLibrary2OpenAI(response *LibraryResponse) *openai.TextRespon
 		FinishReason: "stop",
 	}
 	fullTextResponse := openai.TextResponse{
-		Id:      helper.GetUUID(),
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -66,7 +66,7 @@ func documentsAIProxyLibrary(documents []LibraryDocument) *openai.ChatCompletion
 	choice.Delta.Content = aiProxyDocuments2Markdown(documents)
 	choice.FinishReason = &constant.StopFinishReason
 	return &openai.ChatCompletionsStreamResponse{
-		Id:      helper.GetUUID(),
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "",
@@ -78,7 +78,7 @@ func streamResponseAIProxyLibrary2OpenAI(response *LibraryStreamResponse) *opena
 	var choice openai.ChatCompletionsStreamResponseChoice
 	choice.Delta.Content = response.Content
 	return &openai.ChatCompletionsStreamResponse{
-		Id:      helper.GetUUID(),
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   response.Model,

relay/channel/ali/main.go

@@ -33,6 +33,9 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 		enableSearch = true
 		aliModel = strings.TrimSuffix(aliModel, EnableSearchModelSuffix)
 	}
+	if request.TopP >= 1 {
+		request.TopP = 0.9999
+	}
 	return &ChatRequest{
 		Model: aliModel,
 		Input: Input{
@@ -42,6 +45,9 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 			EnableSearch:      enableSearch,
 			IncrementalOutput: request.Stream,
 			Seed:              uint64(request.Seed),
+			MaxTokens:         request.MaxTokens,
+			Temperature:       request.Temperature,
+			TopP:              request.TopP,
 		},
 	}
 }

relay/channel/ali/model.go

@@ -16,6 +16,8 @@ type Parameters struct {
 	Seed              uint64  `json:"seed,omitempty"`
 	EnableSearch      bool    `json:"enable_search,omitempty"`
 	IncrementalOutput bool    `json:"incremental_output,omitempty"`
+	MaxTokens         int     `json:"max_tokens,omitempty"`
+	Temperature       float64 `json:"temperature,omitempty"`
 }
 
 type ChatRequest struct {

relay/channel/anthropic/adaptor.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/relay/channel"
-	"github.com/songquanpeng/one-api/relay/channel/openai"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
 	"io"
@@ -20,7 +19,7 @@ func (a *Adaptor) Init(meta *util.RelayMeta) {
 }
 
 func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	return fmt.Sprintf("%s/v1/complete", meta.BaseURL), nil
+	return fmt.Sprintf("%s/v1/messages", meta.BaseURL), nil
 }
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
@@ -31,6 +30,7 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *ut
 		anthropicVersion = "2023-06-01"
 	}
 	req.Header.Set("anthropic-version", anthropicVersion)
+	req.Header.Set("anthropic-beta", "messages-2023-12-15")
 	return nil
 }
 
@@ -47,9 +47,7 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
-		var responseText string
-		err, responseText = StreamHandler(c, resp)
-		usage = openai.ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
+		err, usage = StreamHandler(c, resp)
 	} else {
 		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
 	}

relay/channel/anthropic/constants.go

@@ -1,5 +1,8 @@
 package anthropic
 
 var ModelList = []string{
-	"claude-instant-1", "claude-2", "claude-2.0", "claude-2.1",
+	"claude-instant-1.2", "claude-2.0", "claude-2.1",
+	"claude-3-haiku-20240229",
+	"claude-3-sonnet-20240229",
+	"claude-3-opus-20240229",
 }

relay/channel/anthropic/main.go

@@ -7,6 +7,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/helper"
+	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/channel/openai"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -15,73 +16,135 @@ import (
 	"strings"
 )
 
-func stopReasonClaude2OpenAI(reason string) string {
-	switch reason {
+func stopReasonClaude2OpenAI(reason *string) string {
+	if reason == nil {
+		return ""
+	}
+	switch *reason {
+	case "end_turn":
+		return "stop"
 	case "stop_sequence":
 		return "stop"
 	case "max_tokens":
 		return "length"
 	default:
-		return reason
+		return *reason
 	}
 }
 
 func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 	claudeRequest := Request{
-		Model:             textRequest.Model,
-		Prompt:            "",
-		MaxTokensToSample: textRequest.MaxTokens,
-		StopSequences:     nil,
-		Temperature:       textRequest.Temperature,
-		TopP:              textRequest.TopP,
-		Stream:            textRequest.Stream,
+		Model:       textRequest.Model,
+		MaxTokens:   textRequest.MaxTokens,
+		Temperature: textRequest.Temperature,
+		TopP:        textRequest.TopP,
+		Stream:      textRequest.Stream,
 	}
-	if claudeRequest.MaxTokensToSample == 0 {
-		claudeRequest.MaxTokensToSample = 1000000
+	if claudeRequest.MaxTokens == 0 {
+		claudeRequest.MaxTokens = 4096
+	}
+	// legacy model name mapping
+	if claudeRequest.Model == "claude-instant-1" {
+		claudeRequest.Model = "claude-instant-1.1"
+	} else if claudeRequest.Model == "claude-2" {
+		claudeRequest.Model = "claude-2.1"
 	}
-	prompt := ""
 	for _, message := range textRequest.Messages {
-		if message.Role == "user" {
-			prompt += fmt.Sprintf("\n\nHuman: %s", message.Content)
-		} else if message.Role == "assistant" {
-			prompt += fmt.Sprintf("\n\nAssistant: %s", message.Content)
-		} else if message.Role == "system" {
-			if prompt == "" {
-				prompt = message.StringContent()
-			}
+		if message.Role == "system" && claudeRequest.System == "" {
+			claudeRequest.System = message.StringContent()
+			continue
 		}
+		claudeMessage := Message{
+			Role: message.Role,
+		}
+		var content Content
+		if message.IsStringContent() {
+			content.Type = "text"
+			content.Text = message.StringContent()
+			claudeMessage.Content = append(claudeMessage.Content, content)
+			claudeRequest.Messages = append(claudeRequest.Messages, claudeMessage)
+			continue
+		}
+		var contents []Content
+		openaiContent := message.ParseContent()
+		for _, part := range openaiContent {
+			var content Content
+			if part.Type == model.ContentTypeText {
+				content.Type = "text"
+				content.Text = part.Text
+			} else if part.Type == model.ContentTypeImageURL {
+				content.Type = "image"
+				content.Source = &ImageSource{
+					Type: "base64",
+				}
+				mimeType, data, _ := image.GetImageFromUrl(part.ImageURL.Url)
+				content.Source.MediaType = mimeType
+				content.Source.Data = data
+			}
+			contents = append(contents, content)
+		}
+		claudeMessage.Content = contents
+		claudeRequest.Messages = append(claudeRequest.Messages, claudeMessage)
 	}
-	prompt += "\n\nAssistant:"
-	claudeRequest.Prompt = prompt
 	return &claudeRequest
 }
 
-func streamResponseClaude2OpenAI(claudeResponse *Response) *openai.ChatCompletionsStreamResponse {
+// https://docs.anthropic.com/claude/reference/messages-streaming
+func streamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
+	var response *Response
+	var responseText string
+	var stopReason string
+	switch claudeResponse.Type {
+	case "message_start":
+		return nil, claudeResponse.Message
+	case "content_block_start":
+		if claudeResponse.ContentBlock != nil {
+			responseText = claudeResponse.ContentBlock.Text
+		}
+	case "content_block_delta":
+		if claudeResponse.Delta != nil {
+			responseText = claudeResponse.Delta.Text
+		}
+	case "message_delta":
+		if claudeResponse.Usage != nil {
+			response = &Response{
+				Usage: *claudeResponse.Usage,
+			}
+		}
+		if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
+			stopReason = *claudeResponse.Delta.StopReason
+		}
+	}
 	var choice openai.ChatCompletionsStreamResponseChoice
-	choice.Delta.Content = claudeResponse.Completion
-	finishReason := stopReasonClaude2OpenAI(claudeResponse.StopReason)
+	choice.Delta.Content = responseText
+	choice.Delta.Role = "assistant"
+	finishReason := stopReasonClaude2OpenAI(&stopReason)
 	if finishReason != "null" {
 		choice.FinishReason = &finishReason
 	}
-	var response openai.ChatCompletionsStreamResponse
-	response.Object = "chat.completion.chunk"
-	response.Model = claudeResponse.Model
-	response.Choices = []openai.ChatCompletionsStreamResponseChoice{choice}
-	return &response
+	var openaiResponse openai.ChatCompletionsStreamResponse
+	openaiResponse.Object = "chat.completion.chunk"
+	openaiResponse.Choices = []openai.ChatCompletionsStreamResponseChoice{choice}
+	return &openaiResponse, response
 }
 
 func responseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
+	var responseText string
+	if len(claudeResponse.Content) > 0 {
+		responseText = claudeResponse.Content[0].Text
+	}
 	choice := openai.TextResponseChoice{
 		Index: 0,
 		Message: model.Message{
 			Role:    "assistant",
-			Content: strings.TrimPrefix(claudeResponse.Completion, " "),
+			Content: responseText,
 			Name:    nil,
 		},
 		FinishReason: stopReasonClaude2OpenAI(claudeResponse.StopReason),
 	}
 	fullTextResponse := openai.TextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", claudeResponse.Id),
+		Model:   claudeResponse.Model,
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -89,17 +152,15 @@ func responseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
 	return &fullTextResponse
 }
 
-func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, string) {
-	responseText := ""
-	responseId := fmt.Sprintf("chatcmpl-%s", helper.GetUUID())
+func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
 	createdTime := helper.GetTimestamp()
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
 		if atEOF && len(data) == 0 {
 			return 0, nil, nil
 		}
-		if i := strings.Index(string(data), "\r\n\r\n"); i >= 0 {
-			return i + 4, data[0:i], nil
+		if i := strings.Index(string(data), "\n"); i >= 0 {
+			return i + 1, data[0:i], nil
 		}
 		if atEOF {
 			return len(data), data, nil
@@ -111,29 +172,45 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 	go func() {
 		for scanner.Scan() {
 			data := scanner.Text()
-			if !strings.HasPrefix(data, "event: completion") {
+			if len(data) < 6 {
 				continue
 			}
-			data = strings.TrimPrefix(data, "event: completion\r\ndata: ")
+			if !strings.HasPrefix(data, "data: ") {
+				continue
+			}
+			data = strings.TrimPrefix(data, "data: ")
 			dataChan <- data
 		}
 		stopChan <- true
 	}()
 	common.SetEventStreamHeaders(c)
+	var usage model.Usage
+	var modelName string
+	var id string
 	c.Stream(func(w io.Writer) bool {
 		select {
 		case data := <-dataChan:
 			// some implementations may add \r at the end of data
 			data = strings.TrimSuffix(data, "\r")
-			var claudeResponse Response
+			var claudeResponse StreamResponse
 			err := json.Unmarshal([]byte(data), &claudeResponse)
 			if err != nil {
 				logger.SysError("error unmarshalling stream response: " + err.Error())
 				return true
 			}
-			responseText += claudeResponse.Completion
-			response := streamResponseClaude2OpenAI(&claudeResponse)
-			response.Id = responseId
+			response, meta := streamResponseClaude2OpenAI(&claudeResponse)
+			if meta != nil {
+				usage.PromptTokens += meta.Usage.InputTokens
+				usage.CompletionTokens += meta.Usage.OutputTokens
+				modelName = meta.Model
+				id = fmt.Sprintf("chatcmpl-%s", meta.Id)
+				return true
+			}
+			if response == nil {
+				return true
+			}
+			response.Id = id
+			response.Model = modelName
 			response.Created = createdTime
 			jsonStr, err := json.Marshal(response)
 			if err != nil {
@@ -147,11 +224,8 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 			return false
 		}
 	})
-	err := resp.Body.Close()
-	if err != nil {
-		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), ""
-	}
-	return nil, responseText
+	_ = resp.Body.Close()
+	return nil, &usage
 }
 
 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
@@ -181,11 +255,10 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	}
 	fullTextResponse := responseClaude2OpenAI(&claudeResponse)
 	fullTextResponse.Model = modelName
-	completionTokens := openai.CountTokenText(claudeResponse.Completion, modelName)
 	usage := model.Usage{
-		PromptTokens:     promptTokens,
-		CompletionTokens: completionTokens,
-		TotalTokens:      promptTokens + completionTokens,
+		PromptTokens:     claudeResponse.Usage.InputTokens,
+		CompletionTokens: claudeResponse.Usage.OutputTokens,
+		TotalTokens:      claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens,
 	}
 	fullTextResponse.Usage = usage
 	jsonResponse, err := json.Marshal(fullTextResponse)

relay/channel/anthropic/model.go

@@ -1,19 +1,44 @@
 package anthropic
 
+// https://docs.anthropic.com/claude/reference/messages_post
+
 type Metadata struct {
 	UserId string `json:"user_id"`
 }
 
+type ImageSource struct {
+	Type      string `json:"type"`
+	MediaType string `json:"media_type"`
+	Data      string `json:"data"`
+}
+
+type Content struct {
+	Type   string       `json:"type"`
+	Text   string       `json:"text,omitempty"`
+	Source *ImageSource `json:"source,omitempty"`
+}
+
+type Message struct {
+	Role    string    `json:"role"`
+	Content []Content `json:"content"`
+}
+
 type Request struct {
-	Model             string   `json:"model"`
-	Prompt            string   `json:"prompt"`
-	MaxTokensToSample int      `json:"max_tokens_to_sample"`
-	StopSequences     []string `json:"stop_sequences,omitempty"`
-	Temperature       float64  `json:"temperature,omitempty"`
-	TopP              float64  `json:"top_p,omitempty"`
-	TopK              int      `json:"top_k,omitempty"`
+	Model         string    `json:"model"`
+	Messages      []Message `json:"messages"`
+	System        string    `json:"system,omitempty"`
+	MaxTokens     int       `json:"max_tokens,omitempty"`
+	StopSequences []string  `json:"stop_sequences,omitempty"`
+	Stream        bool      `json:"stream,omitempty"`
+	Temperature   float64   `json:"temperature,omitempty"`
+	TopP          float64   `json:"top_p,omitempty"`
+	TopK          int       `json:"top_k,omitempty"`
 	//Metadata    `json:"metadata,omitempty"`
-	Stream bool `json:"stream,omitempty"`
+}
+
+type Usage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
 }
 
 type Error struct {
@@ -22,8 +47,29 @@ type Error struct {
 }
 
 type Response struct {
-	Completion string `json:"completion"`
-	StopReason string `json:"stop_reason"`
-	Model      string `json:"model"`
-	Error      Error  `json:"error"`
+	Id           string    `json:"id"`
+	Type         string    `json:"type"`
+	Role         string    `json:"role"`
+	Content      []Content `json:"content"`
+	Model        string    `json:"model"`
+	StopReason   *string   `json:"stop_reason"`
+	StopSequence *string   `json:"stop_sequence"`
+	Usage        Usage     `json:"usage"`
+	Error        Error     `json:"error"`
+}
+
+type Delta struct {
+	Type         string  `json:"type"`
+	Text         string  `json:"text"`
+	StopReason   *string `json:"stop_reason"`
+	StopSequence *string `json:"stop_sequence"`
+}
+
+type StreamResponse struct {
+	Type         string    `json:"type"`
+	Message      *Response `json:"message"`
+	Index        int       `json:"index"`
+	ContentBlock *Content  `json:"content_block"`
+	Delta        *Delta    `json:"delta"`
+	Usage        *Usage    `json:"usage"`
 }

relay/channel/baichuan/constants.go

@@ -0,0 +1,7 @@
+package baichuan
+
+var ModelList = []string{
+	"Baichuan2-Turbo",
+	"Baichuan2-Turbo-192k",
+	"Baichuan-Text-Embedding",
+}

relay/channel/baidu/adaptor.go

@@ -36,6 +36,8 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 		fullRequestURL = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/bloomz_7b1"
 	case "Embedding-V1":
 		fullRequestURL = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/embeddings/embedding-v1"
+	default:
+               fullRequestURL = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/" + meta.ActualModelName
 	}
 	var accessToken string
 	var err error

relay/channel/gemini/constants.go

@@ -1,6 +1,6 @@
 package gemini
 
 var ModelList = []string{
-	"gemini-pro",
-	"gemini-pro-vision",
+	"gemini-pro", "gemini-1.0-pro-001",
+	"gemini-pro-vision", "gemini-1.0-pro-vision-001",
 }

relay/channel/minimax/constants.go

@@ -0,0 +1,7 @@
+package minimax
+
+var ModelList = []string{
+	"abab5.5s-chat",
+	"abab5.5-chat",
+	"abab6-chat",
+}

relay/channel/minimax/main.go

@@ -0,0 +1,14 @@
+package minimax
+
+import (
+	"fmt"
+	"github.com/songquanpeng/one-api/relay/constant"
+	"github.com/songquanpeng/one-api/relay/util"
+)
+
+func GetRequestURL(meta *util.RelayMeta) (string, error) {
+	if meta.Mode == constant.RelayModeChatCompletions {
+		return fmt.Sprintf("%s/v1/text/chatcompletion_v2", meta.BaseURL), nil
+	}
+	return "", fmt.Errorf("unsupported relay mode %d for minimax", meta.Mode)
+}

relay/channel/mistral/constants.go

@@ -0,0 +1,10 @@
+package mistral
+
+var ModelList = []string{
+	"open-mistral-7b",
+	"open-mixtral-8x7b",
+	"mistral-small-latest",
+	"mistral-medium-latest",
+	"mistral-large-latest",
+	"mistral-embed",
+}

relay/channel/openai/adaptor.go

@@ -7,6 +7,9 @@ import (
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/relay/channel"
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
+	"github.com/songquanpeng/one-api/relay/channel/baichuan"
+	"github.com/songquanpeng/one-api/relay/channel/minimax"
+	"github.com/songquanpeng/one-api/relay/channel/mistral"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
@@ -24,7 +27,8 @@ func (a *Adaptor) Init(meta *util.RelayMeta) {
 }
 
 func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	if meta.ChannelType == common.ChannelTypeAzure {
+	switch meta.ChannelType {
+	case common.ChannelTypeAzure:
 		// https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
 		requestURL := strings.Split(meta.RequestURLPath, "?")[0]
 		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, meta.APIVersion)
@@ -38,8 +42,11 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 
 		requestURL = fmt.Sprintf("/openai/deployments/%s/%s", model_, task)
 		return util.GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType), nil
+	case common.ChannelTypeMinimax:
+		return minimax.GetRequestURL(meta)
+	default:
+		return util.GetFullRequestURL(meta.BaseURL, meta.RequestURLPath, meta.ChannelType), nil
 	}
-	return util.GetFullRequestURL(meta.BaseURL, meta.RequestURLPath, meta.ChannelType), nil
 }
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
@@ -70,7 +77,7 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
 		var responseText string
-		err, responseText = StreamHandler(c, resp, meta.Mode)
+		err, responseText, _ = StreamHandler(c, resp, meta.Mode)
 		usage = ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
 	} else {
 		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
@@ -84,6 +91,12 @@ func (a *Adaptor) GetModelList() []string {
 		return ai360.ModelList
 	case common.ChannelTypeMoonshot:
 		return moonshot.ModelList
+	case common.ChannelTypeBaichuan:
+		return baichuan.ModelList
+	case common.ChannelTypeMinimax:
+		return minimax.ModelList
+	case common.ChannelTypeMistral:
+		return mistral.ModelList
 	default:
 		return ModelList
 	}
@@ -97,6 +110,12 @@ func (a *Adaptor) GetChannelName() string {
 		return "360"
 	case common.ChannelTypeMoonshot:
 		return "moonshot"
+	case common.ChannelTypeBaichuan:
+		return "baichuan"
+	case common.ChannelTypeMinimax:
+		return "minimax"
+	case common.ChannelTypeMistral:
+		return "mistralai"
 	default:
 		return "openai"
 	}

relay/channel/openai/main.go

@@ -14,7 +14,7 @@ import (
 	"strings"
 )
 
-func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string) {
+func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
 	responseText := ""
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
@@ -31,6 +31,7 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 	})
 	dataChan := make(chan string)
 	stopChan := make(chan bool)
+	var usage *model.Usage
 	go func() {
 		for scanner.Scan() {
 			data := scanner.Text()
@@ -54,6 +55,9 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 					for _, choice := range streamResponse.Choices {
 						responseText += choice.Delta.Content
 					}
+					if streamResponse.Usage != nil {
+						usage = streamResponse.Usage
+					}
 				case constant.RelayModeCompletions:
 					var streamResponse CompletionsStreamResponse
 					err := json.Unmarshal([]byte(data), &streamResponse)
@@ -86,9 +90,9 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 	})
 	err := resp.Body.Close()
 	if err != nil {
-		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), ""
+		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", nil
 	}
-	return nil, responseText
+	return nil, responseText, usage
 }
 
 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {

relay/channel/openai/model.go

@@ -118,8 +118,10 @@ type ImageResponse struct {
 }
 
 type ChatCompletionsStreamResponseChoice struct {
+	Index int `json:"index"`
 	Delta struct {
 		Content string `json:"content"`
+		Role    string `json:"role,omitempty"`
 	} `json:"delta"`
 	FinishReason *string `json:"finish_reason,omitempty"`
 }
@@ -130,6 +132,7 @@ type ChatCompletionsStreamResponse struct {
 	Created int64                                 `json:"created"`
 	Model   string                                `json:"model"`
 	Choices []ChatCompletionsStreamResponseChoice `json:"choices"`
+	Usage   *model.Usage                          `json:"usage"`
 }
 
 type CompletionsStreamResponse struct {

relay/channel/tencent/main.go

@@ -81,6 +81,7 @@ func responseTencent2OpenAI(response *ChatResponse) *openai.TextResponse {
 
 func streamResponseTencent2OpenAI(TencentResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
 	response := openai.ChatCompletionsStreamResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "tencent-hunyuan",

relay/channel/xunfei/main.go

@@ -70,6 +70,7 @@ func responseXunfei2OpenAI(response *ChatResponse) *openai.TextResponse {
 		FinishReason: constant.StopFinishReason,
 	}
 	fullTextResponse := openai.TextResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -92,6 +93,7 @@ func streamResponseXunfei2OpenAI(xunfeiResponse *ChatResponse) *openai.ChatCompl
 		choice.FinishReason = &constant.StopFinishReason
 	}
 	response := openai.ChatCompletionsStreamResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "SparkDesk",

relay/channel/zhipu/adaptor.go

@@ -5,20 +5,35 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/relay/channel"
+	"github.com/songquanpeng/one-api/relay/channel/openai"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
+	"strings"
 )
 
 type Adaptor struct {
+	APIVersion string
 }
 
 func (a *Adaptor) Init(meta *util.RelayMeta) {
 
 }
 
+func (a *Adaptor) SetVersionByModeName(modelName string) {
+	if strings.HasPrefix(modelName, "glm-") {
+		a.APIVersion = "v4"
+	} else {
+		a.APIVersion = "v3"
+	}
+}
+
 func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+	a.SetVersionByModeName(meta.ActualModelName)
+	if a.APIVersion == "v4" {
+		return fmt.Sprintf("%s/api/paas/v4/chat/completions", meta.BaseURL), nil
+	}
 	method := "invoke"
 	if meta.IsStream {
 		method = "sse-invoke"
@@ -37,6 +52,13 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
+	if request.TopP >= 1 {
+		request.TopP = 0.99
+	}
+	a.SetVersionByModeName(request.Model)
+	if a.APIVersion == "v4" {
+		return request, nil
+	}
 	return ConvertRequest(*request), nil
 }
 
@@ -44,7 +66,19 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 	return channel.DoRequestHelper(a, c, meta, requestBody)
 }
 
+func (a *Adaptor) DoResponseV4(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if meta.IsStream {
+		err, _, usage = openai.StreamHandler(c, resp, meta.Mode)
+	} else {
+		err, usage = openai.Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	}
+	return
+}
+
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if a.APIVersion == "v4" {
+		return a.DoResponseV4(c, resp, meta)
+	}
 	if meta.IsStream {
 		err, usage = StreamHandler(c, resp)
 	} else {

relay/channel/zhipu/constants.go

@@ -2,4 +2,5 @@ package zhipu
 
 var ModelList = []string{
 	"chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite",
+	"glm-4", "glm-4v", "glm-3-turbo",
 }

relay/constant/image.go

@@ -0,0 +1,24 @@
+package constant
+
+var DalleSizeRatios = map[string]map[string]float64{
+	"dall-e-2": {
+		"256x256":   1,
+		"512x512":   1.125,
+		"1024x1024": 1.25,
+	},
+	"dall-e-3": {
+		"1024x1024": 1,
+		"1024x1792": 2,
+		"1792x1024": 2,
+	},
+}
+
+var DalleGenerationImageAmounts = map[string][2]int{
+	"dall-e-2": {1, 10},
+	"dall-e-3": {1, 1}, // OpenAI allows n=1 currently.
+}
+
+var DalleImagePromptLengthLimitations = map[string]int{
+	"dall-e-2": 1000,
+	"dall-e-3": 4000,
+}

relay/controller/helper.go

@@ -36,6 +36,65 @@ func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.Gener
 	return textRequest, nil
 }
 
+func getImageRequest(c *gin.Context, relayMode int) (*openai.ImageRequest, error) {
+	imageRequest := &openai.ImageRequest{}
+	err := common.UnmarshalBodyReusable(c, imageRequest)
+	if err != nil {
+		return nil, err
+	}
+	if imageRequest.N == 0 {
+		imageRequest.N = 1
+	}
+	if imageRequest.Size == "" {
+		imageRequest.Size = "1024x1024"
+	}
+	if imageRequest.Model == "" {
+		imageRequest.Model = "dall-e-2"
+	}
+	return imageRequest, nil
+}
+
+func validateImageRequest(imageRequest *openai.ImageRequest, meta *util.RelayMeta) *relaymodel.ErrorWithStatusCode {
+	// model validation
+	_, hasValidSize := constant.DalleSizeRatios[imageRequest.Model][imageRequest.Size]
+	if !hasValidSize {
+		return openai.ErrorWrapper(errors.New("size not supported for this image model"), "size_not_supported", http.StatusBadRequest)
+	}
+	// check prompt length
+	if imageRequest.Prompt == "" {
+		return openai.ErrorWrapper(errors.New("prompt is required"), "prompt_missing", http.StatusBadRequest)
+	}
+	if len(imageRequest.Prompt) > constant.DalleImagePromptLengthLimitations[imageRequest.Model] {
+		return openai.ErrorWrapper(errors.New("prompt is too long"), "prompt_too_long", http.StatusBadRequest)
+	}
+	// Number of generated images validation
+	if !isWithinRange(imageRequest.Model, imageRequest.N) {
+		// channel not azure
+		if meta.ChannelType != common.ChannelTypeAzure {
+			return openai.ErrorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
+		}
+	}
+	return nil
+}
+
+func getImageCostRatio(imageRequest *openai.ImageRequest) (float64, error) {
+	if imageRequest == nil {
+		return 0, errors.New("imageRequest is nil")
+	}
+	imageCostRatio, hasValidSize := constant.DalleSizeRatios[imageRequest.Model][imageRequest.Size]
+	if !hasValidSize {
+		return 0, fmt.Errorf("size not supported for this image model: %s", imageRequest.Size)
+	}
+	if imageRequest.Quality == "hd" && imageRequest.Model == "dall-e-3" {
+		if imageRequest.Size == "1024x1024" {
+			imageCostRatio *= 2
+		} else {
+			imageCostRatio *= 1.5
+		}
+	}
+	return imageCostRatio, nil
+}
+
 func getPromptTokens(textRequest *relaymodel.GeneralOpenAIRequest, relayMode int) int {
 	switch relayMode {
 	case constant.RelayModeChatCompletions:
@@ -113,10 +172,8 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *util.R
 	if err != nil {
 		logger.Error(ctx, "error update user quota cache: "+err.Error())
 	}
-	if quota != 0 {
-		logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
-		model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
-		model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
-		model.UpdateChannelUsedQuota(meta.ChannelId, quota)
-	}
+	logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
+	model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
+	model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
+	model.UpdateChannelUsedQuota(meta.ChannelId, quota)
 }

relay/controller/image.go

@@ -10,6 +10,7 @@ import (
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/channel/openai"
+	"github.com/songquanpeng/one-api/relay/constant"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
 	"io"
@@ -20,120 +21,65 @@ import (
 )
 
 func isWithinRange(element string, value int) bool {
-	if _, ok := common.DalleGenerationImageAmounts[element]; !ok {
+	if _, ok := constant.DalleGenerationImageAmounts[element]; !ok {
 		return false
 	}
-	min := common.DalleGenerationImageAmounts[element][0]
-	max := common.DalleGenerationImageAmounts[element][1]
+	min := constant.DalleGenerationImageAmounts[element][0]
+	max := constant.DalleGenerationImageAmounts[element][1]
 
 	return value >= min && value <= max
 }
 
 func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
-	imageModel := "dall-e-2"
-	imageSize := "1024x1024"
-
-	tokenId := c.GetInt("token_id")
-	channelType := c.GetInt("channel")
-	channelId := c.GetInt("channel_id")
-	userId := c.GetInt("id")
-	group := c.GetString("group")
-
-	var imageRequest openai.ImageRequest
-	err := common.UnmarshalBodyReusable(c, &imageRequest)
+	ctx := c.Request.Context()
+	meta := util.GetRelayMeta(c)
+	imageRequest, err := getImageRequest(c, meta.Mode)
 	if err != nil {
-		return openai.ErrorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
-	}
-
-	if imageRequest.N == 0 {
-		imageRequest.N = 1
-	}
-
-	// Size validation
-	if imageRequest.Size != "" {
-		imageSize = imageRequest.Size
-	}
-
-	// Model validation
-	if imageRequest.Model != "" {
-		imageModel = imageRequest.Model
-	}
-
-	imageCostRatio, hasValidSize := common.DalleSizeRatios[imageModel][imageSize]
-
-	// Check if model is supported
-	if hasValidSize {
-		if imageRequest.Quality == "hd" && imageModel == "dall-e-3" {
-			if imageSize == "1024x1024" {
-				imageCostRatio *= 2
-			} else {
-				imageCostRatio *= 1.5
-			}
-		}
-	} else {
-		return openai.ErrorWrapper(errors.New("size not supported for this image model"), "size_not_supported", http.StatusBadRequest)
-	}
-
-	// Prompt validation
-	if imageRequest.Prompt == "" {
-		return openai.ErrorWrapper(errors.New("prompt is required"), "prompt_missing", http.StatusBadRequest)
-	}
-
-	// Check prompt length
-	if len(imageRequest.Prompt) > common.DalleImagePromptLengthLimitations[imageModel] {
-		return openai.ErrorWrapper(errors.New("prompt is too long"), "prompt_too_long", http.StatusBadRequest)
-	}
-
-	// Number of generated images validation
-	if !isWithinRange(imageModel, imageRequest.N) {
-		// channel not azure
-		if channelType != common.ChannelTypeAzure {
-			return openai.ErrorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
-		}
+		logger.Errorf(ctx, "getImageRequest failed: %s", err.Error())
+		return openai.ErrorWrapper(err, "invalid_image_request", http.StatusBadRequest)
 	}
 
 	// map model name
-	modelMapping := c.GetString("model_mapping")
-	isModelMapped := false
-	if modelMapping != "" {
-		modelMap := make(map[string]string)
-		err := json.Unmarshal([]byte(modelMapping), &modelMap)
-		if err != nil {
-			return openai.ErrorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
-		}
-		if modelMap[imageModel] != "" {
-			imageModel = modelMap[imageModel]
-			isModelMapped = true
-		}
+	var isModelMapped bool
+	meta.OriginModelName = imageRequest.Model
+	imageRequest.Model, isModelMapped = util.GetMappedModelName(imageRequest.Model, meta.ModelMapping)
+	meta.ActualModelName = imageRequest.Model
+
+	// model validation
+	bizErr := validateImageRequest(imageRequest, meta)
+	if bizErr != nil {
+		return bizErr
 	}
-	baseURL := common.ChannelBaseURLs[channelType]
+
+	imageCostRatio, err := getImageCostRatio(imageRequest)
+	if err != nil {
+		return openai.ErrorWrapper(err, "get_image_cost_ratio_failed", http.StatusInternalServerError)
+	}
+
 	requestURL := c.Request.URL.String()
-	if c.GetString("base_url") != "" {
-		baseURL = c.GetString("base_url")
-	}
-	fullRequestURL := util.GetFullRequestURL(baseURL, requestURL, channelType)
-	if channelType == common.ChannelTypeAzure {
+	fullRequestURL := util.GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType)
+	if meta.ChannelType == common.ChannelTypeAzure {
 		// https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart?tabs=dalle3%2Ccommand-line&pivots=rest-api
 		apiVersion := util.GetAzureAPIVersion(c)
 		// https://{resource_name}.openai.azure.com/openai/deployments/dall-e-3/images/generations?api-version=2023-06-01-preview
-		fullRequestURL = fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", baseURL, imageModel, apiVersion)
+		fullRequestURL = fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, imageRequest.Model, apiVersion)
 	}
 
 	var requestBody io.Reader
-	if isModelMapped || channelType == common.ChannelTypeAzure { // make Azure channel request body
+	if isModelMapped || meta.ChannelType == common.ChannelTypeAzure { // make Azure channel request body
 		jsonStr, err := json.Marshal(imageRequest)
 		if err != nil {
-			return openai.ErrorWrapper(err, "marshal_text_request_failed", http.StatusInternalServerError)
+			return openai.ErrorWrapper(err, "marshal_image_request_failed", http.StatusInternalServerError)
 		}
 		requestBody = bytes.NewBuffer(jsonStr)
 	} else {
 		requestBody = c.Request.Body
 	}
 
-	modelRatio := common.GetModelRatio(imageModel)
-	groupRatio := common.GetGroupRatio(group)
+	modelRatio := common.GetModelRatio(imageRequest.Model)
+	groupRatio := common.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
-	userQuota, err := model.CacheGetUserQuota(userId)
+	userQuota, err := model.CacheGetUserQuota(meta.UserId)
 
 	quota := int(ratio*imageCostRatio*1000) * imageRequest.N
 
@@ -146,7 +92,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		return openai.ErrorWrapper(err, "new_request_failed", http.StatusInternalServerError)
 	}
 	token := c.Request.Header.Get("Authorization")
-	if channelType == common.ChannelTypeAzure { // Azure authentication
+	if meta.ChannelType == common.ChannelTypeAzure { // Azure authentication
 		token = strings.TrimPrefix(token, "Bearer ")
 		req.Header.Set("api-key", token)
 	} else {
@@ -169,25 +115,25 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
 	}
-	var textResponse openai.ImageResponse
+	var imageResponse openai.ImageResponse
 
 	defer func(ctx context.Context) {
 		if resp.StatusCode != http.StatusOK {
 			return
 		}
-		err := model.PostConsumeTokenQuota(tokenId, quota)
+		err := model.PostConsumeTokenQuota(meta.TokenId, quota)
 		if err != nil {
 			logger.SysError("error consuming token remain quota: " + err.Error())
 		}
-		err = model.CacheUpdateUserQuota(userId)
+		err = model.CacheUpdateUserQuota(meta.UserId)
 		if err != nil {
 			logger.SysError("error update user quota cache: " + err.Error())
 		}
 		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, imageModel, tokenName, quota, logContent)
-			model.UpdateUserUsedQuotaAndRequestCount(userId, quota)
+			model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, 0, 0, imageRequest.Model, tokenName, quota, logContent)
+			model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
 			channelId := c.GetInt("channel_id")
 			model.UpdateChannelUsedQuota(channelId, quota)
 		}
@@ -202,7 +148,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
 	}
-	err = json.Unmarshal(responseBody, &textResponse)
+	err = json.Unmarshal(responseBody, &imageResponse)
 	if err != nil {
 		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
 	}

relay/controller/text.go

@@ -39,6 +39,7 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	ratio := modelRatio * groupRatio
 	// pre-consume quota
 	promptTokens := getPromptTokens(textRequest, meta.Mode)
+	meta.PromptTokens = promptTokens
 	preConsumedQuota, bizErr := preConsumeQuota(ctx, textRequest, promptTokens, ratio, meta)
 	if bizErr != nil {
 		logger.Warnf(ctx, "preConsumeQuota failed: %+v", *bizErr)
@@ -54,7 +55,8 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	var requestBody io.Reader
 	if meta.APIType == constant.APITypeOpenAI {
 		// no need to convert request for openai
-		if isModelMapped {
+		shouldResetRequestBody := isModelMapped || meta.ChannelType == common.ChannelTypeBaichuan // 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)

web/THEMES

@@ -1,2 +1,2 @@
 default
-berry
+berry

web/berry/src/constants/ChannelConstants.js

@@ -29,6 +29,12 @@ export const CHANNEL_OPTIONS = {
     value: 24,
     color: 'orange'
   },
+  28: {
+    key: 28,
+    text: 'Mistral AI',
+    value: 28,
+    color: 'orange'
+  },
   15: {
     key: 15,
     text: '百度文心千帆',
@@ -71,6 +77,18 @@ export const CHANNEL_OPTIONS = {
     value: 23,
     color: 'default'
   },
+  26: {
+    key: 26,
+    text: '百川大模型',
+    value: 26,
+    color: 'default'
+  },
+  27: {
+    key: 27,
+    text: 'MiniMax',
+    value: 27,
+    color: 'default'
+  },
   8: {
     key: 8,
     text: '自定义渠道',

web/berry/src/views/Channel/type/Config.js

@@ -67,7 +67,7 @@ const typeConfig = {
   },
   16: {
     input: {
-      models: ["chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite"],
+      models: ["glm-4", "glm-4v", "glm-3-turbo", "chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite"],
     },
     modelGroup: "zhipu",
   },
@@ -145,6 +145,24 @@ const typeConfig = {
     },
     modelGroup: "google gemini",
   },
+  25: {
+    input: {
+      models: ['moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k'],
+    },
+    modelGroup: "moonshot",
+  },
+  26: {
+    input: {
+      models: ['Baichuan2-Turbo', 'Baichuan2-Turbo-192k', 'Baichuan-Text-Embedding'],
+    },
+    modelGroup: "baichuan",
+  },
+  27: {
+    input: {
+      models: ['abab5.5s-chat', 'abab5.5-chat', 'abab6-chat'],
+    },
+    modelGroup: "minimax",
+  },
 };
 
 export { defaultConfig, typeConfig };

web/build.sh

@@ -1,13 +1,13 @@
 #!/bin/sh
 
 version=$(cat VERSION)
-themes=$(cat THEMES)
-IFS=$'\n'
+pwd
 
-for theme in $themes; do
+while IFS= read -r theme; do
     echo "Building theme: $theme"
-    cd $theme
+    rm -r build/$theme
+    cd "$theme"
     npm install
     DISABLE_ESLINT_PLUGIN='true' REACT_APP_VERSION=$version npm run build
     cd ..
-done
+done < THEMES

web/default/src/constants/channel.constants.js

@@ -4,6 +4,7 @@ export const CHANNEL_OPTIONS = [
   { key: 3, text: 'Azure OpenAI', value: 3, color: 'olive' },
   { key: 11, text: 'Google PaLM2', value: 11, color: 'orange' },
   { key: 24, text: 'Google Gemini', value: 24, color: 'orange' },
+  { key: 28, text: 'Mistral AI', value: 28, color: 'orange' },
   { key: 15, text: '百度文心千帆', value: 15, color: 'blue' },
   { key: 17, text: '阿里通义千问', value: 17, color: 'orange' },
   { key: 18, text: '讯飞星火认知', value: 18, color: 'blue' },
@@ -11,6 +12,8 @@ export const CHANNEL_OPTIONS = [
   { key: 19, text: '360 智脑', value: 19, color: 'blue' },
   { key: 25, text: 'Moonshot AI', value: 25, color: 'black' },
   { key: 23, text: '腾讯混元', value: 23, color: 'teal' },
+  { key: 26, text: '百川大模型', value: 26, color: 'orange' },
+  { key: 27, text: 'MiniMax', value: 27, color: 'red' },
   { key: 8, text: '自定义渠道', value: 8, color: 'pink' },
   { key: 22, text: '知识库：FastGPT', value: 22, color: 'blue' },
   { key: 21, text: '知识库：AI Proxy', value: 21, color: 'purple' },

web/default/src/pages/Channel/EditChannel.js

@@ -79,7 +79,7 @@ const EditChannel = () => {
           localModels = [...localModels, ...withInternetVersion];
           break;
         case 16:
-          localModels = ['chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
+          localModels = ["glm-4", "glm-4v", "glm-3-turbo",'chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
           break;
         case 18:
           localModels = [
@@ -102,6 +102,12 @@ const EditChannel = () => {
         case 25:
           localModels = ['moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k'];
           break;
+        case 26:
+          localModels = ['Baichuan2-Turbo', 'Baichuan2-Turbo-192k', 'Baichuan-Text-Embedding'];
+          break;
+        case 27:
+          localModels = ['abab5.5s-chat', 'abab5.5-chat', 'abab6-chat'];
+          break;
       }
       setInputs((inputs) => ({ ...inputs, models: localModels }));
     }