Merge branch 'main' into patch/images-edits

README.md

@@ -84,6 +84,9 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [阶跃星辰](https://platform.stepfun.com/)
    + [x] [Coze](https://www.coze.com/)
    + [x] [Cohere](https://cohere.com/)
+   + [x] [DeepSeek](https://www.deepseek.com/)
+   + [x] [Cloudflare Workers AI](https://developers.cloudflare.com/workers-ai/)
+   + [x] [DeepL](https://www.deepl.com/)
 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/ctxkey/config.go

@@ -1,13 +0,0 @@
-package ctxkey
-
-const (
-	ConfigPrefix = "cfg_"
-
-	ConfigAPIVersion = ConfigPrefix + "api_version"
-	ConfigLibraryID  = ConfigPrefix + "library_id"
-	ConfigPlugin     = ConfigPrefix + "plugin"
-	ConfigSK         = ConfigPrefix + "sk"
-	ConfigAK         = ConfigPrefix + "ak"
-	ConfigRegion     = ConfigPrefix + "region"
-	ConfigUserID     = ConfigPrefix + "user_id"
-)

common/ctxkey/key.go

@@ -1,6 +1,7 @@
 package ctxkey
 
 const (
+	Config            = "config"
 	Id                = "id"
 	RequestId         = "X-Oneapi-Request-Id"
 	Username          = "username"

common/helper/helper.go

@@ -2,6 +2,7 @@ package helper
 
 import (
 	"fmt"
+	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/random"
 	"html/template"
 	"log"
@@ -105,6 +106,11 @@ func GenRequestID() string {
 	return GetTimeString() + random.GetRandomNumberString(8)
 }
 
+func GetResponseID(c *gin.Context) string {
+	logID := c.GetString(RequestIdKey)
+	return fmt.Sprintf("chatcmpl-%s", logID)
+}
+
 func Max(a int, b int) int {
 	if a >= b {
 		return a

common/helper/key.go

@@ -0,0 +1,5 @@
+package helper
+
+const (
+	RequestIdKey = "X-Oneapi-Request-Id"
+)

common/logger/logger.go

@@ -88,7 +88,7 @@ func logHelper(ctx context.Context, level string, msg string) {
 	if level == loggerINFO {
 		writer = gin.DefaultWriter
 	}
-	id := ctx.Value(ctxkey.RequestId)
+	id := ctx.Value(helper.RequestIdKey)
 	if id == nil {
 		id = helper.GenRequestID()
 	}

controller/channel-test.go

@@ -5,6 +5,15 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/common/logger"
@@ -18,14 +27,6 @@ import (
 	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
-	"io"
-	"net/http"
-	"net/http/httptest"
-	"net/url"
-	"strconv"
-	"strings"
-	"sync"
-	"time"
 
 	"github.com/gin-gonic/gin"
 )
@@ -57,6 +58,8 @@ func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error
 	c.Request.Header.Set("Content-Type", "application/json")
 	c.Set(ctxkey.Channel, channel.Type)
 	c.Set(ctxkey.BaseURL, channel.GetBaseURL())
+	cfg, _ := channel.LoadConfig()
+	c.Set(ctxkey.Config, cfg)
 	middleware.SetupContextForSelectedChannel(c, channel, "")
 	meta := meta.GetByContext(c)
 	apiType := channeltype.ToAPIType(channel.Type)
@@ -67,6 +70,7 @@ func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error
 	adaptor.Init(meta)
 	var modelName string
 	modelList := adaptor.GetModelList()
+	modelMap := channel.GetModelMapping()
 	if len(modelList) != 0 {
 		modelName = modelList[0]
 	}
@@ -75,6 +79,9 @@ func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error
 		if len(modelNames) > 0 {
 			modelName = modelNames[0]
 		}
+		if modelMap != nil && modelMap[modelName] != "" {
+			modelName = modelMap[modelName]
+		}
 	}
 	request := buildTestRequest()
 	request.Model = modelName

middleware/distributor.go

@@ -66,21 +66,29 @@ func SetupContextForSelectedChannel(c *gin.Context, channel *model.Channel, mode
 	c.Set(ctxkey.OriginalModel, modelName) // for retry
 	c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))
 	c.Set(ctxkey.BaseURL, channel.GetBaseURL())
+	cfg, _ := channel.LoadConfig()
 	// this is for backward compatibility
 	switch channel.Type {
 	case channeltype.Azure:
-		c.Set(ctxkey.ConfigAPIVersion, channel.Other)
+		if cfg.APIVersion == "" {
+			cfg.APIVersion = channel.Other
+		}
 	case channeltype.Xunfei:
-		c.Set(ctxkey.ConfigAPIVersion, channel.Other)
+		if cfg.APIVersion == "" {
+			cfg.APIVersion = channel.Other
+		}
 	case channeltype.Gemini:
-		c.Set(ctxkey.ConfigAPIVersion, channel.Other)
+		if cfg.APIVersion == "" {
+			cfg.APIVersion = channel.Other
+		}
 	case channeltype.AIProxyLibrary:
-		c.Set(ctxkey.ConfigLibraryID, channel.Other)
+		if cfg.LibraryID == "" {
+			cfg.LibraryID = channel.Other
+		}
 	case channeltype.Ali:
-		c.Set(ctxkey.ConfigPlugin, channel.Other)
+		if cfg.Plugin == "" {
-	}
+			cfg.Plugin = channel.Other
-	cfg, _ := channel.LoadConfig()
-	for k, v := range cfg {
-		c.Set(ctxkey.ConfigPrefix+k, v)
 		}
 	}
+	c.Set(ctxkey.Config, cfg)
+}

middleware/logger.go

@@ -4,14 +4,14 @@ import (
 	"fmt"
 
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/ctxkey"
+	"github.com/songquanpeng/one-api/common/helper"
 )
 
 func SetUpLogger(server *gin.Engine) {
 	server.Use(gin.LoggerWithFormatter(func(param gin.LogFormatterParams) string {
 		var requestID string
 		if param.Keys != nil {
-			requestID = param.Keys[ctxkey.RequestId].(string)
+			requestID = param.Keys[helper.RequestIdKey].(string)
 		}
 		return fmt.Sprintf("[GIN] %s | %s | %3d | %13v | %15s | %7s %s\n",
 			param.TimeStamp.Format("2006/01/02 - 15:04:05"),

middleware/request-id.go

@@ -11,10 +11,10 @@ import (
 func RequestId() func(c *gin.Context) {
 	return func(c *gin.Context) {
 		id := helper.GenRequestID()
-		c.Set(ctxkey.RequestId, id)
+		c.Set(helper.RequestIdKey, id)
-		ctx := context.WithValue(c.Request.Context(), ctxkey.RequestId, id)
+		ctx := context.WithValue(c.Request.Context(), helper.RequestIdKey, id)
 		c.Request = c.Request.WithContext(ctx)
-		c.Header(ctxkey.RequestId, id)
+		c.Header(helper.RequestIdKey, id)
 		c.Next()
 	}
 }

middleware/utils.go

@@ -14,7 +14,7 @@ import (
 func abortWithMessage(c *gin.Context, statusCode int, message string) {
 	c.JSON(statusCode, gin.H{
 		"error": gin.H{
-			"message": helper.MessageWithRequestId(message, c.GetString(ctxkey.RequestId)),
+			"message": helper.MessageWithRequestId(message, c.GetString(helper.RequestIdKey)),
 			"type":    "one_api_error",
 		},
 	})

model/channel.go

@@ -38,6 +38,16 @@ type Channel struct {
 	Config             string  `json:"config"`
 }
 
+type ChannelConfig struct {
+	Region     string `json:"region,omitempty"`
+	SK         string `json:"sk,omitempty"`
+	AK         string `json:"ak,omitempty"`
+	UserID     string `json:"user_id,omitempty"`
+	APIVersion string `json:"api_version,omitempty"`
+	LibraryID  string `json:"library_id,omitempty"`
+	Plugin     string `json:"plugin,omitempty"`
+}
+
 func GetAllChannels(startIdx int, num int, scope string) ([]*Channel, error) {
 	var channels []*Channel
 	var err error
@@ -161,14 +171,14 @@ func (channel *Channel) Delete() error {
 	return err
 }
 
-func (channel *Channel) LoadConfig() (map[string]string, error) {
+func (channel *Channel) LoadConfig() (ChannelConfig, error) {
+	var cfg ChannelConfig
 	if channel.Config == "" {
-		return nil, nil
+		return cfg, nil
 	}
-	cfg := make(map[string]string)
 	err := json.Unmarshal([]byte(channel.Config), &cfg)
 	if err != nil {
-		return nil, err
+		return cfg, err
 	}
 	return cfg, nil
 }

relay/adaptor.go

@@ -7,8 +7,10 @@ import (
 	"github.com/songquanpeng/one-api/relay/adaptor/anthropic"
 	"github.com/songquanpeng/one-api/relay/adaptor/aws"
 	"github.com/songquanpeng/one-api/relay/adaptor/baidu"
+	"github.com/songquanpeng/one-api/relay/adaptor/cloudflare"
 	"github.com/songquanpeng/one-api/relay/adaptor/cohere"
 	"github.com/songquanpeng/one-api/relay/adaptor/coze"
+	"github.com/songquanpeng/one-api/relay/adaptor/deepl"
 	"github.com/songquanpeng/one-api/relay/adaptor/gemini"
 	"github.com/songquanpeng/one-api/relay/adaptor/ollama"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
@@ -49,6 +51,10 @@ func GetAdaptor(apiType int) adaptor.Adaptor {
 		return &coze.Adaptor{}
 	case apitype.Cohere:
 		return &cohere.Adaptor{}
+	case apitype.Cloudflare:
+		return &cloudflare.Adaptor{}
+	case apitype.DeepL:
+		return &deepl.Adaptor{}
 	}
 	return nil
 }

relay/adaptor/aiproxy/adaptor.go

@@ -4,7 +4,6 @@ import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -13,10 +12,11 @@ import (
 )
 
 type Adaptor struct {
+	meta *meta.Meta
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
+	a.meta = meta
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
@@ -34,7 +34,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		return nil, errors.New("request is nil")
 	}
 	aiProxyLibraryRequest := ConvertRequest(*request)
-	aiProxyLibraryRequest.LibraryId = c.GetString(ctxkey.ConfigLibraryID)
+	aiProxyLibraryRequest.LibraryId = a.meta.Config.LibraryID
 	return aiProxyLibraryRequest, nil
 }
 

relay/adaptor/ali/adaptor.go

@@ -4,7 +4,6 @@ import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -16,10 +15,11 @@ import (
 // https://help.aliyun.com/zh/dashscope/developer-reference/api-details
 
 type Adaptor struct {
+	meta *meta.Meta
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
+	a.meta = meta
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
@@ -47,8 +47,8 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 	if meta.Mode == relaymode.ImagesGenerations {
 		req.Header.Set("X-DashScope-Async", "enable")
 	}
-	if c.GetString(ctxkey.ConfigPlugin) != "" {
+	if a.meta.Config.Plugin != "" {
-		req.Header.Set("X-DashScope-Plugin", c.GetString(ctxkey.ConfigPlugin))
+		req.Header.Set("X-DashScope-Plugin", a.meta.Config.Plugin)
 	}
 	return nil
 }

relay/adaptor/anthropic/main.go

@@ -4,6 +4,10 @@ import (
 	"bufio"
 	"encoding/json"
 	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/helper"
@@ -11,9 +15,6 @@ import (
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/model"
-	"io"
-	"net/http"
-	"strings"
 )
 
 func stopReasonClaude2OpenAI(reason *string) string {
@@ -192,7 +193,7 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 		select {
 		case data := <-dataChan:
 			// some implementations may add \r at the end of data
-			data = strings.TrimSuffix(data, "\r")
+			data = strings.TrimSpace(data)
 			var claudeResponse StreamResponse
 			err := json.Unmarshal([]byte(data), &claudeResponse)
 			if err != nil {

relay/adaptor/aws/adapter.go

@@ -1,6 +1,9 @@
 package aws
 
 import (
+	"github.com/aws/aws-sdk-go-v2/aws"
+	"github.com/aws/aws-sdk-go-v2/credentials"
+	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"io"
 	"net/http"
@@ -16,10 +19,16 @@ import (
 var _ adaptor.Adaptor = new(Adaptor)
 
 type Adaptor struct {
+	meta      *meta.Meta
+	awsClient *bedrockruntime.Client
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
+	a.meta = meta
+	a.awsClient = bedrockruntime.New(bedrockruntime.Options{
+		Region:      meta.Config.Region,
+		Credentials: aws.NewCredentialsCache(credentials.NewStaticCredentialsProvider(meta.Config.AK, meta.Config.SK, "")),
+	})
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
@@ -54,9 +63,9 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Read
 
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
-		err, usage = StreamHandler(c, resp)
+		err, usage = StreamHandler(c, a.awsClient)
 	} else {
-		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+		err, usage = Handler(c, a.awsClient, meta.ActualModelName)
 	}
 	return
 }
@@ -65,7 +74,6 @@ func (a *Adaptor) GetModelList() (models []string) {
 	for n := range awsModelIDMap {
 		models = append(models, n)
 	}
-
 	return
 }
 

relay/adaptor/aws/main.go

@@ -10,7 +10,6 @@ import (
 	"net/http"
 
 	"github.com/aws/aws-sdk-go-v2/aws"
-	"github.com/aws/aws-sdk-go-v2/credentials"
 	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
 	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime/types"
 	"github.com/gin-gonic/gin"
@@ -23,18 +22,6 @@ import (
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 )
 
-func newAwsClient(c *gin.Context) (*bedrockruntime.Client, error) {
-	ak := c.GetString(ctxkey.ConfigAK)
-	sk := c.GetString(ctxkey.ConfigSK)
-	region := c.GetString(ctxkey.ConfigRegion)
-	client := bedrockruntime.New(bedrockruntime.Options{
-		Region:      region,
-		Credentials: aws.NewCredentialsCache(credentials.NewStaticCredentialsProvider(ak, sk, "")),
-	})
-
-	return client, nil
-}
-
 func wrapErr(err error) *relaymodel.ErrorWithStatusCode {
 	return &relaymodel.ErrorWithStatusCode{
 		StatusCode: http.StatusInternalServerError,
@@ -62,12 +49,7 @@ func awsModelID(requestModel string) (string, error) {
 	return "", errors.Errorf("model %s not found", requestModel)
 }
 
-func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
+func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
-	awsCli, err := newAwsClient(c)
-	if err != nil {
-		return wrapErr(errors.Wrap(err, "newAwsClient")), nil
-	}
-
 	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return wrapErr(errors.Wrap(err, "awsModelID")), nil
@@ -120,13 +102,8 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	return nil, &usage
 }
 
-func StreamHandler(c *gin.Context, resp *http.Response) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
+func StreamHandler(c *gin.Context, awsCli *bedrockruntime.Client) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
 	createdTime := helper.GetTimestamp()
-	awsCli, err := newAwsClient(c)
-	if err != nil {
-		return wrapErr(errors.Wrap(err, "newAwsClient")), nil
-	}
-
 	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return wrapErr(errors.Wrap(err, "awsModelID")), nil

relay/adaptor/azure/helper.go

@@ -1,15 +0,0 @@
-package azure
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/ctxkey"
-)
-
-func GetAPIVersion(c *gin.Context) string {
-	query := c.Request.URL.Query()
-	apiVersion := query.Get("api-version")
-	if apiVersion == "" {
-		apiVersion = c.GetString(ctxkey.ConfigAPIVersion)
-	}
-	return apiVersion
-}

relay/adaptor/cloudflare/adaptor.go

@@ -0,0 +1,66 @@
+package cloudflare
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+type Adaptor struct {
+	meta *meta.Meta
+}
+
+// ConvertImageRequest implements adaptor.Adaptor.
+func (*Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+	return nil, errors.New("not implemented")
+}
+
+// ConvertImageRequest implements adaptor.Adaptor.
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.meta = meta
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	return fmt.Sprintf("%s/client/v4/accounts/%s/ai/run/%s", meta.BaseURL, meta.Config.UserID, meta.ActualModelName), nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
+	req.Header.Set("Authorization", "Bearer "+meta.APIKey)
+	return nil
+}
+
+func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	return ConvertRequest(*request), nil
+}
+
+func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error) {
+	return adaptor.DoRequestHelper(a, c, meta, requestBody)
+}
+
+func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if meta.IsStream {
+		err, usage = StreamHandler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	} else {
+		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "cloudflare"
+}

relay/adaptor/cloudflare/constant.go

@@ -0,0 +1,36 @@
+package cloudflare
+
+var ModelList = []string{
+	"@cf/meta/llama-2-7b-chat-fp16",
+	"@cf/meta/llama-2-7b-chat-int8",
+	"@cf/mistral/mistral-7b-instruct-v0.1",
+	"@hf/thebloke/deepseek-coder-6.7b-base-awq",
+	"@hf/thebloke/deepseek-coder-6.7b-instruct-awq",
+	"@cf/deepseek-ai/deepseek-math-7b-base",
+	"@cf/deepseek-ai/deepseek-math-7b-instruct",
+	"@cf/thebloke/discolm-german-7b-v1-awq",
+	"@cf/tiiuae/falcon-7b-instruct",
+	"@cf/google/gemma-2b-it-lora",
+	"@hf/google/gemma-7b-it",
+	"@cf/google/gemma-7b-it-lora",
+	"@hf/nousresearch/hermes-2-pro-mistral-7b",
+	"@hf/thebloke/llama-2-13b-chat-awq",
+	"@cf/meta-llama/llama-2-7b-chat-hf-lora",
+	"@cf/meta/llama-3-8b-instruct",
+	"@hf/thebloke/llamaguard-7b-awq",
+	"@hf/thebloke/mistral-7b-instruct-v0.1-awq",
+	"@hf/mistralai/mistral-7b-instruct-v0.2",
+	"@cf/mistral/mistral-7b-instruct-v0.2-lora",
+	"@hf/thebloke/neural-chat-7b-v3-1-awq",
+	"@cf/openchat/openchat-3.5-0106",
+	"@hf/thebloke/openhermes-2.5-mistral-7b-awq",
+	"@cf/microsoft/phi-2",
+	"@cf/qwen/qwen1.5-0.5b-chat",
+	"@cf/qwen/qwen1.5-1.8b-chat",
+	"@cf/qwen/qwen1.5-14b-chat-awq",
+	"@cf/qwen/qwen1.5-7b-chat-awq",
+	"@cf/defog/sqlcoder-7b-2",
+	"@hf/nexusflow/starling-lm-7b-beta",
+	"@cf/tinyllama/tinyllama-1.1b-chat-v1.0",
+	"@hf/thebloke/zephyr-7b-beta-awq",
+}

relay/adaptor/cloudflare/main.go

@@ -0,0 +1,152 @@
+package cloudflare
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"io"
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/helper"
+	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
+	lastMessage := textRequest.Messages[len(textRequest.Messages)-1]
+	return &Request{
+		MaxTokens:   textRequest.MaxTokens,
+		Prompt:      lastMessage.StringContent(),
+		Stream:      textRequest.Stream,
+		Temperature: textRequest.Temperature,
+	}
+}
+
+func ResponseCloudflare2OpenAI(cloudflareResponse *Response) *openai.TextResponse {
+	choice := openai.TextResponseChoice{
+		Index: 0,
+		Message: model.Message{
+			Role:    "assistant",
+			Content: cloudflareResponse.Result.Response,
+		},
+		FinishReason: "stop",
+	}
+	fullTextResponse := openai.TextResponse{
+		Object:  "chat.completion",
+		Created: helper.GetTimestamp(),
+		Choices: []openai.TextResponseChoice{choice},
+	}
+	return &fullTextResponse
+}
+
+func StreamResponseCloudflare2OpenAI(cloudflareResponse *StreamResponse) *openai.ChatCompletionsStreamResponse {
+	var choice openai.ChatCompletionsStreamResponseChoice
+	choice.Delta.Content = cloudflareResponse.Response
+	choice.Delta.Role = "assistant"
+	openaiResponse := openai.ChatCompletionsStreamResponse{
+		Object:  "chat.completion.chunk",
+		Choices: []openai.ChatCompletionsStreamResponseChoice{choice},
+		Created: helper.GetTimestamp(),
+	}
+	return &openaiResponse
+}
+
+func StreamHandler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
+	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 := bytes.IndexByte(data, '\n'); i >= 0 {
+			return i + 1, data[0:i], nil
+		}
+		if atEOF {
+			return len(data), data, nil
+		}
+		return 0, nil, nil
+	})
+
+	dataChan := make(chan string)
+	stopChan := make(chan bool)
+	go func() {
+		for scanner.Scan() {
+			data := scanner.Text()
+			if len(data) < len("data: ") {
+				continue
+			}
+			data = strings.TrimPrefix(data, "data: ")
+			dataChan <- data
+		}
+		stopChan <- true
+	}()
+	common.SetEventStreamHeaders(c)
+	id := helper.GetResponseID(c)
+	responseModel := c.GetString("original_model")
+	var responseText 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 cloudflareResponse StreamResponse
+			err := json.Unmarshal([]byte(data), &cloudflareResponse)
+			if err != nil {
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return true
+			}
+			response := StreamResponseCloudflare2OpenAI(&cloudflareResponse)
+			if response == nil {
+				return true
+			}
+			responseText += cloudflareResponse.Response
+			response.Id = id
+			response.Model = responseModel
+			jsonStr, err := json.Marshal(response)
+			if err != nil {
+				logger.SysError("error marshalling stream response: " + err.Error())
+				return true
+			}
+			c.Render(-1, common.CustomEvent{Data: "data: " + string(jsonStr)})
+			return true
+		case <-stopChan:
+			c.Render(-1, common.CustomEvent{Data: "data: [DONE]"})
+			return false
+		}
+	})
+	_ = resp.Body.Close()
+	usage := openai.ResponseText2Usage(responseText, responseModel, promptTokens)
+	return nil, usage
+}
+
+func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+	var cloudflareResponse Response
+	err = json.Unmarshal(responseBody, &cloudflareResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	fullTextResponse := ResponseCloudflare2OpenAI(&cloudflareResponse)
+	fullTextResponse.Model = modelName
+	usage := openai.ResponseText2Usage(cloudflareResponse.Result.Response, modelName, promptTokens)
+	fullTextResponse.Usage = *usage
+	fullTextResponse.Id = helper.GetResponseID(c)
+	jsonResponse, err := json.Marshal(fullTextResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(jsonResponse)
+	return nil, usage
+}

relay/adaptor/cloudflare/model.go

@@ -0,0 +1,25 @@
+package cloudflare
+
+type Request struct {
+	Lora        string  `json:"lora,omitempty"`
+	MaxTokens   int     `json:"max_tokens,omitempty"`
+	Prompt      string  `json:"prompt,omitempty"`
+	Raw         bool    `json:"raw,omitempty"`
+	Stream      bool    `json:"stream,omitempty"`
+	Temperature float64 `json:"temperature,omitempty"`
+}
+
+type Result struct {
+	Response string `json:"response"`
+}
+
+type Response struct {
+	Result   Result   `json:"result"`
+	Success  bool     `json:"success"`
+	Errors   []string `json:"errors"`
+	Messages []string `json:"messages"`
+}
+
+type StreamResponse struct {
+	Response string `json:"response"`
+}

relay/adaptor/cohere/constant.go

@@ -5,3 +5,10 @@ var ModelList = []string{
 	"command-light", "command-light-nightly",
 	"command-r", "command-r-plus",
 }
+
+func init() {
+	num := len(ModelList)
+	for i := 0; i < num; i++ {
+		ModelList = append(ModelList, ModelList[i]+"-internet")
+	}
+}

relay/adaptor/cohere/main.go

@@ -17,6 +17,10 @@ import (
 	"github.com/songquanpeng/one-api/relay/model"
 )
 
+var (
+	WebSearchConnector = Connector{ID: "web-search"}
+)
+
 func stopReasonCohere2OpenAI(reason *string) string {
 	if reason == nil {
 		return ""
@@ -45,6 +49,10 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 	if cohereRequest.Model == "" {
 		cohereRequest.Model = "command-r"
 	}
+	if strings.HasSuffix(cohereRequest.Model, "-internet") {
+		cohereRequest.Model = strings.TrimSuffix(cohereRequest.Model, "-internet")
+		cohereRequest.Connectors = append(cohereRequest.Connectors, WebSearchConnector)
+	}
 	for _, message := range textRequest.Messages {
 		if message.Role == "user" {
 			cohereRequest.Message = message.Content.(string)

relay/adaptor/coze/adaptor.go

@@ -4,7 +4,6 @@ import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/meta"
@@ -14,10 +13,11 @@ import (
 )
 
 type Adaptor struct {
+	meta *meta.Meta
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
+	a.meta = meta
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
@@ -34,7 +34,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
-	request.User = c.GetString(ctxkey.ConfigUserID)
+	request.User = a.meta.Config.UserID
 	return ConvertRequest(*request), nil
 }
 

relay/adaptor/deepl/adaptor.go

@@ -0,0 +1,73 @@
+package deepl
+
+import (
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
+)
+
+type Adaptor struct {
+	meta       *meta.Meta
+	promptText string
+}
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.meta = meta
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	return fmt.Sprintf("%s/v2/translate", meta.BaseURL), nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
+	req.Header.Set("Authorization", "DeepL-Auth-Key "+meta.APIKey)
+	return nil
+}
+
+func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	convertedRequest, text := ConvertRequest(*request)
+	a.promptText = text
+	return convertedRequest, nil
+}
+
+func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	return request, nil
+}
+
+func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error) {
+	return adaptor.DoRequestHelper(a, c, meta, requestBody)
+}
+
+func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if meta.IsStream {
+		err = StreamHandler(c, resp, meta.ActualModelName)
+	} else {
+		err = Handler(c, resp, meta.ActualModelName)
+	}
+	promptTokens := len(a.promptText)
+	usage = &model.Usage{
+		PromptTokens: promptTokens,
+		TotalTokens:  promptTokens,
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "deepl"
+}

relay/adaptor/deepl/constants.go

@@ -0,0 +1,9 @@
+package deepl
+
+// https://developers.deepl.com/docs/api-reference/glossaries
+
+var ModelList = []string{
+	"deepl-zh",
+	"deepl-en",
+	"deepl-ja",
+}

relay/adaptor/deepl/helper.go

@@ -0,0 +1,11 @@
+package deepl
+
+import "strings"
+
+func parseLangFromModelName(modelName string) string {
+	parts := strings.Split(modelName, "-")
+	if len(parts) == 1 {
+		return "ZH"
+	}
+	return parts[1]
+}

relay/adaptor/deepl/main.go

@@ -0,0 +1,137 @@
+package deepl
+
+import (
+	"encoding/json"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/helper"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/constant"
+	"github.com/songquanpeng/one-api/relay/constant/finishreason"
+	"github.com/songquanpeng/one-api/relay/constant/role"
+	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
+)
+
+// https://developers.deepl.com/docs/getting-started/your-first-api-request
+
+func ConvertRequest(textRequest model.GeneralOpenAIRequest) (*Request, string) {
+	var text string
+	if len(textRequest.Messages) != 0 {
+		text = textRequest.Messages[len(textRequest.Messages)-1].StringContent()
+	}
+	deeplRequest := Request{
+		TargetLang: parseLangFromModelName(textRequest.Model),
+		Text:       []string{text},
+	}
+	return &deeplRequest, text
+}
+
+func StreamResponseDeepL2OpenAI(deeplResponse *Response) *openai.ChatCompletionsStreamResponse {
+	var choice openai.ChatCompletionsStreamResponseChoice
+	if len(deeplResponse.Translations) != 0 {
+		choice.Delta.Content = deeplResponse.Translations[0].Text
+	}
+	choice.Delta.Role = role.Assistant
+	choice.FinishReason = &constant.StopFinishReason
+	openaiResponse := openai.ChatCompletionsStreamResponse{
+		Object:  constant.StreamObject,
+		Created: helper.GetTimestamp(),
+		Choices: []openai.ChatCompletionsStreamResponseChoice{choice},
+	}
+	return &openaiResponse
+}
+
+func ResponseDeepL2OpenAI(deeplResponse *Response) *openai.TextResponse {
+	var responseText string
+	if len(deeplResponse.Translations) != 0 {
+		responseText = deeplResponse.Translations[0].Text
+	}
+	choice := openai.TextResponseChoice{
+		Index: 0,
+		Message: model.Message{
+			Role:    role.Assistant,
+			Content: responseText,
+			Name:    nil,
+		},
+		FinishReason: finishreason.Stop,
+	}
+	fullTextResponse := openai.TextResponse{
+		Object:  constant.NonStreamObject,
+		Created: helper.GetTimestamp(),
+		Choices: []openai.TextResponseChoice{choice},
+	}
+	return &fullTextResponse
+}
+
+func StreamHandler(c *gin.Context, resp *http.Response, modelName string) *model.ErrorWithStatusCode {
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	var deeplResponse Response
+	err = json.Unmarshal(responseBody, &deeplResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+	}
+	fullTextResponse := StreamResponseDeepL2OpenAI(&deeplResponse)
+	fullTextResponse.Model = modelName
+	fullTextResponse.Id = helper.GetResponseID(c)
+	jsonData, err := json.Marshal(fullTextResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError)
+	}
+	common.SetEventStreamHeaders(c)
+	c.Stream(func(w io.Writer) bool {
+		if jsonData != nil {
+			c.Render(-1, common.CustomEvent{Data: "data: " + string(jsonData)})
+			jsonData = nil
+			return true
+		}
+		c.Render(-1, common.CustomEvent{Data: "data: [DONE]"})
+		return false
+	})
+	_ = resp.Body.Close()
+	return nil
+}
+
+func Handler(c *gin.Context, resp *http.Response, modelName string) *model.ErrorWithStatusCode {
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	var deeplResponse Response
+	err = json.Unmarshal(responseBody, &deeplResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+	}
+	if deeplResponse.Message != "" {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: deeplResponse.Message,
+				Code:    "deepl_error",
+			},
+			StatusCode: resp.StatusCode,
+		}
+	}
+	fullTextResponse := ResponseDeepL2OpenAI(&deeplResponse)
+	fullTextResponse.Model = modelName
+	fullTextResponse.Id = helper.GetResponseID(c)
+	jsonResponse, err := json.Marshal(fullTextResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError)
+	}
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(jsonResponse)
+	return nil
+}

relay/adaptor/deepl/model.go

@@ -0,0 +1,16 @@
+package deepl
+
+type Request struct {
+	Text       []string `json:"text"`
+	TargetLang string   `json:"target_lang"`
+}
+
+type Translation struct {
+	DetectedSourceLanguage string `json:"detected_source_language,omitempty"`
+	Text                   string `json:"text,omitempty"`
+}
+
+type Response struct {
+	Translations []Translation `json:"translations,omitempty"`
+	Message      string        `json:"message,omitempty"`
+}

relay/adaptor/deepseek/constants.go

@@ -0,0 +1,6 @@
+package deepseek
+
+var ModelList = []string{
+	"deepseek-chat",
+	"deepseek-coder",
+}

relay/adaptor/gemini/adaptor.go

@@ -3,6 +3,9 @@ package gemini
 import (
 	"errors"
 	"fmt"
+	"io"
+	"net/http"
+
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
@@ -10,8 +13,6 @@ import (
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"io"
-	"net/http"
 )
 
 type Adaptor struct {
@@ -22,10 +23,10 @@ func (a *Adaptor) Init(meta *meta.Meta) {
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
-	version := helper.AssignOrDefault(meta.APIVersion, config.GeminiVersion)
+	version := helper.AssignOrDefault(meta.Config.APIVersion, config.GeminiVersion)
 	action := "generateContent"
 	if meta.IsStream {
-		action = "streamGenerateContent"
+		action = "streamGenerateContent?alt=sse"
 	}
 	return fmt.Sprintf("%s/%s/models/%s:%s", meta.BaseURL, version, meta.ActualModelName, action), nil
 }

relay/adaptor/gemini/main.go

@@ -232,8 +232,6 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *ChatResponse) *openai.ChatC
 
 func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, string) {
 	responseText := ""
-	dataChan := make(chan string)
-	stopChan := make(chan bool)
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
 		if atEOF && len(data) == 0 {
@@ -247,14 +245,16 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 		}
 		return 0, nil, nil
 	})
+	dataChan := make(chan string)
+	stopChan := make(chan bool)
 	go func() {
 		for scanner.Scan() {
 			data := scanner.Text()
 			data = strings.TrimSpace(data)
-			if !strings.HasPrefix(data, "\"text\": \"") {
+			if !strings.HasPrefix(data, "data: ") {
 				continue
 			}
-			data = strings.TrimPrefix(data, "\"text\": \"")
+			data = strings.TrimPrefix(data, "data: ")
 			data = strings.TrimSuffix(data, "\"")
 			dataChan <- data
 		}
@@ -264,23 +264,17 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 	c.Stream(func(w io.Writer) bool {
 		select {
 		case data := <-dataChan:
-			// this is used to prevent annoying \ related format bug
+			var geminiResponse ChatResponse
-			data = fmt.Sprintf("{\"content\": \"%s\"}", data)
+			err := json.Unmarshal([]byte(data), &geminiResponse)
-			type dummyStruct struct {
+			if err != nil {
-				Content string `json:"content"`
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return true
 			}
-			var dummy dummyStruct
+			response := streamResponseGeminiChat2OpenAI(&geminiResponse)
-			err := json.Unmarshal([]byte(data), &dummy)
+			if response == nil {
-			responseText += dummy.Content
+				return true
-			var choice openai.ChatCompletionsStreamResponseChoice
-			choice.Delta.Content = dummy.Content
-			response := openai.ChatCompletionsStreamResponse{
-				Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
-				Object:  "chat.completion.chunk",
-				Created: helper.GetTimestamp(),
-				Model:   "gemini-pro",
-				Choices: []openai.ChatCompletionsStreamResponseChoice{choice},
 			}
+			responseText += response.Choices[0].Delta.StringContent()
 			jsonResponse, err := json.Marshal(response)
 			if err != nil {
 				logger.SysError("error marshalling stream response: " + err.Error())

relay/adaptor/ollama/constants.go

@@ -1,5 +1,11 @@
 package ollama
 
 var ModelList = []string{
+	"codellama:7b-instruct",
+	"llama2:7b",
+	"llama2:latest",
+	"llama3:latest",
+	"phi3:latest",
 	"qwen:0.5b-chat",
+	"qwen:7b",
 }

relay/adaptor/ollama/main.go

@@ -13,6 +13,7 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/constant"
@@ -32,9 +33,22 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 		Stream: request.Stream,
 	}
 	for _, message := range request.Messages {
+		openaiContent := message.ParseContent()
+		var imageUrls []string
+		var contentText string
+		for _, part := range openaiContent {
+			switch part.Type {
+			case model.ContentTypeText:
+				contentText = part.Text
+			case model.ContentTypeImageURL:
+				_, data, _ := image.GetImageFromUrl(part.ImageURL.Url)
+				imageUrls = append(imageUrls, data)
+			}
+		}
 		ollamaRequest.Messages = append(ollamaRequest.Messages, Message{
 			Role:    message.Role,
-			Content: message.StringContent(),
+			Content: contentText,
+			Images:  imageUrls,
 		})
 	}
 	return &ollamaRequest

relay/adaptor/openai/adaptor.go

@@ -29,13 +29,13 @@ func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 		if meta.Mode == relaymode.ImagesGenerations {
 			// https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart?tabs=dalle3%2Ccommand-line&pivots=rest-api
 			// https://{resource_name}.openai.azure.com/openai/deployments/dall-e-3/images/generations?api-version=2024-03-01-preview
-			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, meta.APIVersion)
+			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, meta.Config.APIVersion)
 			return fullRequestURL, nil
 		}
 
 		// 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)
+		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, meta.Config.APIVersion)
 		task := strings.TrimPrefix(requestURL, "/v1/")
 		model_ := meta.ActualModelName
 		model_ = strings.Replace(model_, ".", "", -1)

relay/adaptor/openai/compatible.go

@@ -3,6 +3,7 @@ package openai
 import (
 	"github.com/songquanpeng/one-api/relay/adaptor/ai360"
 	"github.com/songquanpeng/one-api/relay/adaptor/baichuan"
+	"github.com/songquanpeng/one-api/relay/adaptor/deepseek"
 	"github.com/songquanpeng/one-api/relay/adaptor/groq"
 	"github.com/songquanpeng/one-api/relay/adaptor/lingyiwanwu"
 	"github.com/songquanpeng/one-api/relay/adaptor/minimax"
@@ -22,6 +23,7 @@ var CompatibleChannels = []int{
 	channeltype.Groq,
 	channeltype.LingYiWanWu,
 	channeltype.StepFun,
+	channeltype.DeepSeek,
 }
 
 func GetCompatibleChannelMeta(channelType int) (string, []string) {
@@ -44,6 +46,8 @@ func GetCompatibleChannelMeta(channelType int) (string, []string) {
 		return "lingyiwanwu", lingyiwanwu.ModelList
 	case channeltype.StepFun:
 		return "stepfun", stepfun.ModelList
+	case channeltype.DeepSeek:
+		return "deepseek", deepseek.ModelList
 	default:
 		return "openai", ModelList
 	}

relay/adaptor/openai/model.go

@@ -134,7 +134,7 @@ type ChatCompletionsStreamResponse struct {
 	Created int64                                 `json:"created"`
 	Model   string                                `json:"model"`
 	Choices []ChatCompletionsStreamResponseChoice `json:"choices"`
-	Usage   *model.Usage                          `json:"usage"`
+	Usage   *model.Usage                          `json:"usage,omitempty"`
 }
 
 type CompletionsStreamResponse struct {

relay/adaptor/openai/token.go

@@ -206,3 +206,7 @@ func CountTokenText(text string, model string) int {
 	tokenEncoder := getTokenEncoder(model)
 	return getTokenNum(tokenEncoder, text)
 }
+
+func CountToken(text string) int {
+	return CountTokenInput(text, "gpt-3.5-turbo")
+}

relay/adaptor/xunfei/adaptor.go

@@ -14,10 +14,11 @@ import (
 
 type Adaptor struct {
 	request *model.GeneralOpenAIRequest
+	meta    *meta.Meta
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
+	a.meta = meta
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
@@ -26,6 +27,14 @@ func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
 	adaptor.SetupCommonRequestHeader(c, req, meta)
+	version := parseAPIVersionByModelName(meta.ActualModelName)
+	if version == "" {
+		version = a.meta.Config.APIVersion
+	}
+	if version == "" {
+		version = "v1.1"
+	}
+	a.meta.Config.APIVersion = version
 	// check DoResponse for auth part
 	return nil
 }
@@ -61,9 +70,9 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Met
 		return nil, openai.ErrorWrapper(errors.New("request is nil"), "request_is_nil", http.StatusBadRequest)
 	}
 	if meta.IsStream {
-		err, usage = StreamHandler(c, *a.request, splits[0], splits[1], splits[2])
+		err, usage = StreamHandler(c, meta, *a.request, splits[0], splits[1], splits[2])
 	} else {
-		err, usage = Handler(c, *a.request, splits[0], splits[1], splits[2])
+		err, usage = Handler(c, meta, *a.request, splits[0], splits[1], splits[2])
 	}
 	return
 }

relay/adaptor/xunfei/main.go

@@ -9,12 +9,12 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/gorilla/websocket"
 	"github.com/songquanpeng/one-api/common"
-	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/common/random"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/constant"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
 	"io"
 	"net/http"
@@ -149,8 +149,8 @@ func buildXunfeiAuthUrl(hostUrl string, apiKey, apiSecret string) string {
 	return callUrl
 }
 
-func StreamHandler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
+func StreamHandler(c *gin.Context, meta *meta.Meta, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
-	domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model)
+	domain, authUrl := getXunfeiAuthUrl(meta.Config.APIVersion, apiKey, apiSecret)
 	dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId)
 	if err != nil {
 		return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil
@@ -179,8 +179,8 @@ func StreamHandler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId
 	return nil, &usage
 }
 
-func Handler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
+func Handler(c *gin.Context, meta *meta.Meta, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
-	domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model)
+	domain, authUrl := getXunfeiAuthUrl(meta.Config.APIVersion, apiKey, apiSecret)
 	dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId)
 	if err != nil {
 		return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil
@@ -268,25 +268,12 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl,
 	return dataChan, stopChan, nil
 }
 
-func getAPIVersion(c *gin.Context, modelName string) string {
+func parseAPIVersionByModelName(modelName string) string {
-	query := c.Request.URL.Query()
-	apiVersion := query.Get("api-version")
-	if apiVersion != "" {
-		return apiVersion
-	}
 	parts := strings.Split(modelName, "-")
 	if len(parts) == 2 {
-		apiVersion = parts[1]
+		return parts[1]
-		return apiVersion
-
 	}
-	apiVersion = c.GetString(ctxkey.ConfigAPIVersion)
+	return ""
-	if apiVersion != "" {
-		return apiVersion
-	}
-	apiVersion = "v1.1"
-	logger.SysLog("api_version not found, using default: " + apiVersion)
-	return apiVersion
 }
 
 // https://www.xfyun.cn/doc/spark/Web.html#_1-%E6%8E%A5%E5%8F%A3%E8%AF%B4%E6%98%8E
@@ -304,8 +291,7 @@ func apiVersion2domain(apiVersion string) string {
 	return "general" + apiVersion
 }
 
-func getXunfeiAuthUrl(c *gin.Context, apiKey string, apiSecret string, modelName string) (string, string) {
+func getXunfeiAuthUrl(apiVersion string, apiKey string, apiSecret string) (string, string) {
-	apiVersion := getAPIVersion(c, modelName)
 	domain := apiVersion2domain(apiVersion)
 	authUrl := buildXunfeiAuthUrl(fmt.Sprintf("wss://spark-api.xf-yun.com/%s/chat", apiVersion), apiKey, apiSecret)
 	return domain, authUrl

relay/adaptor/zhipu/adaptor.go

@@ -62,8 +62,8 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 	switch relayMode {
 	case relaymode.Embeddings:
-		baiduEmbeddingRequest := ConvertEmbeddingRequest(*request)
+		baiduEmbeddingRequest, err := ConvertEmbeddingRequest(*request)
-		return baiduEmbeddingRequest, nil
+		return baiduEmbeddingRequest, err
 	default:
 		// TopP (0.0, 1.0)
 		request.TopP = math.Min(0.99, request.TopP)
@@ -129,11 +129,15 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Met
 	return
 }
 
-func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) *EmbeddingRequest {
+func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) (*EmbeddingRequest, error) {
-	return &EmbeddingRequest{
+	inputs := request.ParseInput()
-		Model: "embedding-2",
+	if len(inputs) != 1 {
-		Input: request.Input.(string),
+		return nil, errors.New("invalid input length, zhipu only support one input")
 	}
+	return &EmbeddingRequest{
+		Model: request.Model,
+		Input: inputs[0],
+	}, nil
 }
 
 func (a *Adaptor) GetModelList() []string {

relay/apitype/define.go

@@ -15,6 +15,8 @@ const (
 	AwsClaude
 	Coze
 	Cohere
+	Cloudflare
+	DeepL
 
 	Dummy // this one is only for count, do not add any channel after this
 )

relay/billing/ratio/model.go

@@ -170,6 +170,13 @@ var ModelRatio = map[string]float64{
 	"command-light-nightly": 0.5,
 	"command-r":             0.5 / 1000 * USD,
 	"command-r-plus	":       3.0 / 1000 * USD,
+	// https://platform.deepseek.com/api-docs/pricing/
+	"deepseek-chat":  1.0 / 1000 * RMB,
+	"deepseek-coder": 1.0 / 1000 * RMB,
+	// https://www.deepl.com/pro?cta=header-prices
+	"deepl-zh": 25.0 / 1000 * USD,
+	"deepl-en": 25.0 / 1000 * USD,
+	"deepl-ja": 25.0 / 1000 * USD,
 }
 
 var CompletionRatio = map[string]float64{}
@@ -225,6 +232,9 @@ func GetModelRatio(name string) float64 {
 	if strings.HasPrefix(name, "qwen-") && strings.HasSuffix(name, "-internet") {
 		name = strings.TrimSuffix(name, "-internet")
 	}
+	if strings.HasPrefix(name, "command-") && strings.HasSuffix(name, "-internet") {
+		name = strings.TrimSuffix(name, "-internet")
+	}
 	ratio, ok := ModelRatio[name]
 	if !ok {
 		ratio, ok = DefaultModelRatio[name]
@@ -285,6 +295,9 @@ func GetCompletionRatio(name string) float64 {
 	if strings.HasPrefix(name, "gemini-") {
 		return 3
 	}
+	if strings.HasPrefix(name, "deepseek-") {
+		return 2
+	}
 	switch name {
 	case "llama2-70b-4096":
 		return 0.8 / 0.64

relay/channeltype/define.go

@@ -37,6 +37,9 @@ const (
 	AwsClaude
 	Coze
 	Cohere
+	DeepSeek
+	Cloudflare
+	DeepL
 
 	Dummy
 )

relay/channeltype/helper.go

@@ -31,6 +31,10 @@ func ToAPIType(channelType int) int {
 		apiType = apitype.Coze
 	case Cohere:
 		apiType = apitype.Cohere
+	case Cloudflare:
+		apiType = apitype.Cloudflare
+	case DeepL:
+		apiType = apitype.DeepL
 	}
 
 	return apiType

relay/channeltype/url.go

@@ -37,6 +37,9 @@ var ChannelBaseURLs = []string{
 	"",                                          // 33
 	"https://api.coze.com",                      // 34
 	"https://api.cohere.ai",                     // 35
+	"https://api.deepseek.com",                  // 36
+	"https://api.cloudflare.com",                // 37
+	"https://api-free.deepl.com",                // 38
 }
 
 func init() {

relay/constant/common.go

@@ -1,3 +1,5 @@
 package constant
 
 var StopFinishReason = "stop"
+var StreamObject = "chat.completion.chunk"
+var NonStreamObject = "chat.completion"

relay/constant/finishreason/define.go

@@ -0,0 +1,5 @@
+package finishreason
+
+const (
+	Stop = "stop"
+)

relay/constant/role/define.go

@@ -0,0 +1,5 @@
+package role
+
+const (
+	Assistant = "assistant"
+)

relay/controller/audio.go

@@ -13,12 +13,12 @@ import (
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/model"
-	"github.com/songquanpeng/one-api/relay/adaptor/azure"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/billing"
 	billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/client"
+	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 	"io"
@@ -28,6 +28,7 @@ import (
 
 func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
 	ctx := c.Request.Context()
+	meta := meta.GetByContext(c)
 	audioModel := "whisper-1"
 
 	tokenId := c.GetInt(ctxkey.TokenId)
@@ -128,7 +129,7 @@ func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 
 	fullRequestURL := openai.GetFullRequestURL(baseURL, requestURL, channelType)
 	if channelType == channeltype.Azure {
-		apiVersion := azure.GetAPIVersion(c)
+		apiVersion := meta.Config.APIVersion
 		if relayMode == relaymode.AudioTranscription {
 			// https://learn.microsoft.com/en-us/azure/ai-services/openai/whisper-quickstart?tabs=command-line#rest-api
 			fullRequestURL = fmt.Sprintf("%s/openai/deployments/%s/audio/transcriptions?api-version=%s", baseURL, audioModel, apiVersion)

relay/controller/helper.go

@@ -18,6 +18,7 @@ import (
 	"github.com/songquanpeng/one-api/relay/relaymode"
 	"math"
 	"net/http"
+	"strings"
 )
 
 func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.GeneralOpenAIRequest, error) {
@@ -124,9 +125,9 @@ func getPromptTokens(textRequest *relaymodel.GeneralOpenAIRequest, relayMode int
 }
 
 func getPreConsumedQuota(textRequest *relaymodel.GeneralOpenAIRequest, promptTokens int, ratio float64) int64 {
-	preConsumedTokens := config.PreConsumedQuota
+	preConsumedTokens := config.PreConsumedQuota + int64(promptTokens)
 	if textRequest.MaxTokens != 0 {
-		preConsumedTokens = int64(promptTokens) + int64(textRequest.MaxTokens)
+		preConsumedTokens += int64(textRequest.MaxTokens)
 	}
 	return int64(float64(preConsumedTokens) * ratio)
 }
@@ -204,3 +205,20 @@ func getMappedModelName(modelName string, mapping map[string]string) (string, bo
 	}
 	return modelName, false
 }
+
+func isErrorHappened(meta *meta.Meta, resp *http.Response) bool {
+	if resp == nil {
+		return true
+	}
+	if resp.StatusCode != http.StatusOK {
+		return true
+	}
+	if meta.ChannelType == channeltype.DeepL {
+		// skip stream check for deepl
+		return false
+	}
+	if meta.IsStream && strings.HasPrefix(resp.Header.Get("Content-Type"), "application/json") {
+		return true
+	}
+	return false
+}

relay/controller/image.go

@@ -73,6 +73,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 	if adaptor == nil {
 		return openai.ErrorWrapper(fmt.Errorf("invalid api type: %d", meta.APIType), "invalid_api_type", http.StatusBadRequest)
 	}
+	adaptor.Init(meta)
 
 	switch meta.ChannelType {
 	case channeltype.Ali:

relay/controller/text.go

@@ -4,10 +4,6 @@ import (
 	"bytes"
 	"encoding/json"
 	"fmt"
-	"io"
-	"net/http"
-	"strings"
-
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay"
@@ -18,6 +14,8 @@ import (
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
 )
 
 func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
@@ -53,6 +51,7 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	if adaptor == nil {
 		return openai.ErrorWrapper(fmt.Errorf("invalid api type: %d", meta.APIType), "invalid_api_type", http.StatusBadRequest)
 	}
+	adaptor.Init(meta)
 
 	// get request body
 	var requestBody io.Reader
@@ -87,13 +86,10 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 		logger.Errorf(ctx, "DoRequest failed: %s", err.Error())
 		return openai.ErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
 	}
-	if resp != nil {
+	if isErrorHappened(meta, resp) {
-		errorHappened := (resp.StatusCode != http.StatusOK) || (meta.IsStream && strings.HasPrefix(resp.Header.Get("Content-Type"), "application/json"))
-		if errorHappened {
 		billing.ReturnPreConsumedQuota(ctx, preConsumedQuota, meta.TokenId)
 		return RelayErrorHandler(resp)
 	}
-	}
 
 	// do response
 	usage, respErr := adaptor.DoResponse(c, resp, meta)

relay/meta/relay_meta.go

@@ -3,7 +3,7 @@ package meta
 import (
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/ctxkey"
-	"github.com/songquanpeng/one-api/relay/adaptor/azure"
+	"github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 	"strings"
@@ -19,10 +19,9 @@ type Meta struct {
 	Group           string
 	ModelMapping    map[string]string
 	BaseURL         string
-	APIVersion      string
 	APIKey          string
 	APIType         int
-	Config          map[string]string
+	Config          model.ChannelConfig
 	IsStream        bool
 	OriginModelName string
 	ActualModelName string
@@ -40,14 +39,14 @@ func GetByContext(c *gin.Context) *Meta {
 		UserId:          c.GetInt(ctxkey.Id),
 		Group:           c.GetString(ctxkey.Group),
 		ModelMapping:    c.GetStringMapString(ctxkey.ModelMapping),
+		OriginModelName: c.GetString(ctxkey.RequestModel),
 		BaseURL:         c.GetString(ctxkey.BaseURL),
-		APIVersion:     c.GetString(ctxkey.ConfigAPIVersion),
 		APIKey:          strings.TrimPrefix(c.Request.Header.Get("Authorization"), "Bearer "),
-		Config:         nil,
 		RequestURLPath:  c.Request.URL.String(),
 	}
-	if meta.ChannelType == channeltype.Azure {
+	cfg, ok := c.Get(ctxkey.Config)
-		meta.APIVersion = azure.GetAPIVersion(c)
+	if ok {
+		meta.Config = cfg.(model.ChannelConfig)
 	}
 	if meta.BaseURL == "" {
 		meta.BaseURL = channeltype.ChannelBaseURLs[meta.ChannelType]

web/berry/src/constants/ChannelConstants.js

@@ -131,6 +131,18 @@ export const CHANNEL_OPTIONS = {
     value: 35,
     color: 'primary'
   },
+  36: {
+    key: 36,
+    text: 'DeepSeek',
+    value: 36,
+    color: 'primary'
+  },
+  38: {
+    key: 38,
+    text: 'DeepL',
+    value: 38,
+    color: 'primary'
+  },
   8: {
     key: 8,
     text: '自定义渠道',

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

@@ -21,6 +21,9 @@ export const CHANNEL_OPTIONS = [
     {key: 32, text: '阶跃星辰', value: 32, color: 'blue'},
     {key: 34, text: 'Coze', value: 34, color: 'blue'},
     {key: 35, text: 'Cohere', value: 35, color: 'blue'},
+    {key: 36, text: 'DeepSeek', value: 36, color: 'black'},
+    {key: 37, text: 'Cloudflare', value: 37, color: 'orange'},
+    {key: 38, text: 'DeepL', value: 38, color: 'black'},
     {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

@@ -488,6 +488,21 @@ const EditChannel = () => {
               />
             </Form.Field>)
           }
+          {
+            inputs.type === 37 && (
+              <Form.Field>
+                <Form.Input
+                  label='Account ID'
+                  name='user_id'
+                  required
+                  placeholder={'请输入 Account ID，例如：d8d7c61dbc334c32d3ced580e4bf42b4'}
+                  onChange={handleConfigChange}
+                  value={config.user_id}
+                  autoComplete=''
+                />
+              </Form.Field>
+            )
+          }
           {
             inputs.type !== 33 && !isEdit && (
               <Form.Checkbox