Merge branch 'main' into customer-function

relay/adaptor.go

@@ -0,0 +1,60 @@
+package relay
+
+import (
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/aiproxy"
+	"github.com/songquanpeng/one-api/relay/adaptor/ali"
+	"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"
+	"github.com/songquanpeng/one-api/relay/adaptor/palm"
+	"github.com/songquanpeng/one-api/relay/adaptor/tencent"
+	"github.com/songquanpeng/one-api/relay/adaptor/xunfei"
+	"github.com/songquanpeng/one-api/relay/adaptor/zhipu"
+	"github.com/songquanpeng/one-api/relay/apitype"
+)
+
+func GetAdaptor(apiType int) adaptor.Adaptor {
+	switch apiType {
+	case apitype.AIProxyLibrary:
+		return &aiproxy.Adaptor{}
+	case apitype.Ali:
+		return &ali.Adaptor{}
+	case apitype.Anthropic:
+		return &anthropic.Adaptor{}
+	case apitype.AwsClaude:
+		return &aws.Adaptor{}
+	case apitype.Baidu:
+		return &baidu.Adaptor{}
+	case apitype.Gemini:
+		return &gemini.Adaptor{}
+	case apitype.OpenAI:
+		return &openai.Adaptor{}
+	case apitype.PaLM:
+		return &palm.Adaptor{}
+	case apitype.Tencent:
+		return &tencent.Adaptor{}
+	case apitype.Xunfei:
+		return &xunfei.Adaptor{}
+	case apitype.Zhipu:
+		return &zhipu.Adaptor{}
+	case apitype.Ollama:
+		return &ollama.Adaptor{}
+	case apitype.Coze:
+		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

@@ -4,27 +4,27 @@ import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common"
-	"github.com/songquanpeng/one-api/relay/channel"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 )
 
 type Adaptor struct {
+	meta *meta.Meta
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
-
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.meta = meta
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	return fmt.Sprintf("%s/api/library/ask", meta.BaseURL), nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+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
 }
@@ -34,15 +34,22 @@ 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(common.ConfigKeyLibraryID)
+	aiProxyLibraryRequest.LibraryId = a.meta.Config.LibraryID
 	return aiProxyLibraryRequest, nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
+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) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+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)
 	} else {

relay/adaptor/aiproxy/constants.go

@@ -1,6 +1,6 @@
 package aiproxy
 
-import "github.com/songquanpeng/one-api/relay/channel/openai"
+import "github.com/songquanpeng/one-api/relay/adaptor/openai"
 
 var ModelList = []string{""}
 

relay/adaptor/aiproxy/main.go

@@ -8,7 +8,8 @@ import (
 	"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/channel/openai"
+	"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/model"
 	"io"
@@ -53,7 +54,7 @@ func responseAIProxyLibrary2OpenAI(response *LibraryResponse) *openai.TextRespon
 		FinishReason: "stop",
 	}
 	fullTextResponse := openai.TextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -66,7 +67,7 @@ func documentsAIProxyLibrary(documents []LibraryDocument) *openai.ChatCompletion
 	choice.Delta.Content = aiProxyDocuments2Markdown(documents)
 	choice.FinishReason = &constant.StopFinishReason
 	return &openai.ChatCompletionsStreamResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "",
@@ -78,7 +79,7 @@ func streamResponseAIProxyLibrary2OpenAI(response *LibraryStreamResponse) *opena
 	var choice openai.ChatCompletionsStreamResponseChoice
 	choice.Delta.Content = response.Content
 	return &openai.ChatCompletionsStreamResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   response.Model,

relay/adaptor/ali/adaptor.go

@@ -0,0 +1,105 @@
+package ali
+
+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"
+	"github.com/songquanpeng/one-api/relay/relaymode"
+	"io"
+	"net/http"
+)
+
+// 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) {
+	fullRequestURL := ""
+	switch meta.Mode {
+	case relaymode.Embeddings:
+		fullRequestURL = fmt.Sprintf("%s/api/v1/services/embeddings/text-embedding/text-embedding", meta.BaseURL)
+	case relaymode.ImagesGenerations:
+		fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/text2image/image-synthesis", meta.BaseURL)
+	default:
+		fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/text-generation/generation", meta.BaseURL)
+	}
+
+	return fullRequestURL, nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
+	if meta.IsStream {
+		req.Header.Set("Accept", "text/event-stream")
+		req.Header.Set("X-DashScope-SSE", "enable")
+	}
+	req.Header.Set("Authorization", "Bearer "+meta.APIKey)
+
+	if meta.Mode == relaymode.ImagesGenerations {
+		req.Header.Set("X-DashScope-Async", "enable")
+	}
+	if a.meta.Config.Plugin != "" {
+		req.Header.Set("X-DashScope-Plugin", a.meta.Config.Plugin)
+	}
+	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")
+	}
+	switch relayMode {
+	case relaymode.Embeddings:
+		aliEmbeddingRequest := ConvertEmbeddingRequest(*request)
+		return aliEmbeddingRequest, nil
+	default:
+		aliRequest := ConvertRequest(*request)
+		return aliRequest, nil
+	}
+}
+
+func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+
+	aliRequest := ConvertImageRequest(*request)
+	return aliRequest, 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)
+	} else {
+		switch meta.Mode {
+		case relaymode.Embeddings:
+			err, usage = EmbeddingHandler(c, resp)
+		case relaymode.ImagesGenerations:
+			err, usage = ImageHandler(c, resp)
+		default:
+			err, usage = Handler(c, resp)
+		}
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "ali"
+}

relay/adaptor/ali/constants.go

@@ -3,4 +3,5 @@ package ali
 var ModelList = []string{
 	"qwen-turbo", "qwen-plus", "qwen-max", "qwen-max-longcontext",
 	"text-embedding-v1",
+	"ali-stable-diffusion-xl", "ali-stable-diffusion-v1.5", "wanx-v1",
 }

relay/adaptor/ali/image.go

@@ -0,0 +1,192 @@
+package ali
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"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"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+func ImageHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	apiKey := c.Request.Header.Get("Authorization")
+	apiKey = strings.TrimPrefix(apiKey, "Bearer ")
+	responseFormat := c.GetString("response_format")
+
+	var aliTaskResponse TaskResponse
+	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
+	}
+	err = json.Unmarshal(responseBody, &aliTaskResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	if aliTaskResponse.Message != "" {
+		logger.SysError("aliAsyncTask err: " + string(responseBody))
+		return openai.ErrorWrapper(errors.New(aliTaskResponse.Message), "ali_async_task_failed", http.StatusInternalServerError), nil
+	}
+
+	aliResponse, _, err := asyncTaskWait(aliTaskResponse.Output.TaskId, apiKey)
+	if err != nil {
+		return openai.ErrorWrapper(err, "ali_async_task_wait_failed", http.StatusInternalServerError), nil
+	}
+
+	if aliResponse.Output.TaskStatus != "SUCCEEDED" {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: aliResponse.Output.Message,
+				Type:    "ali_error",
+				Param:   "",
+				Code:    aliResponse.Output.Code,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+
+	fullTextResponse := responseAli2OpenAIImage(aliResponse, responseFormat)
+	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, nil
+}
+
+func asyncTask(taskID string, key string) (*TaskResponse, error, []byte) {
+	url := fmt.Sprintf("https://dashscope.aliyuncs.com/api/v1/tasks/%s", taskID)
+
+	var aliResponse TaskResponse
+
+	req, err := http.NewRequest("GET", url, nil)
+	if err != nil {
+		return &aliResponse, err, nil
+	}
+
+	req.Header.Set("Authorization", "Bearer "+key)
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		logger.SysError("aliAsyncTask client.Do err: " + err.Error())
+		return &aliResponse, err, nil
+	}
+	defer resp.Body.Close()
+
+	responseBody, err := io.ReadAll(resp.Body)
+
+	var response TaskResponse
+	err = json.Unmarshal(responseBody, &response)
+	if err != nil {
+		logger.SysError("aliAsyncTask NewDecoder err: " + err.Error())
+		return &aliResponse, err, nil
+	}
+
+	return &response, nil, responseBody
+}
+
+func asyncTaskWait(taskID string, key string) (*TaskResponse, []byte, error) {
+	waitSeconds := 2
+	step := 0
+	maxStep := 20
+
+	var taskResponse TaskResponse
+	var responseBody []byte
+
+	for {
+		step++
+		rsp, err, body := asyncTask(taskID, key)
+		responseBody = body
+		if err != nil {
+			return &taskResponse, responseBody, err
+		}
+
+		if rsp.Output.TaskStatus == "" {
+			return &taskResponse, responseBody, nil
+		}
+
+		switch rsp.Output.TaskStatus {
+		case "FAILED":
+			fallthrough
+		case "CANCELED":
+			fallthrough
+		case "SUCCEEDED":
+			fallthrough
+		case "UNKNOWN":
+			return rsp, responseBody, nil
+		}
+		if step >= maxStep {
+			break
+		}
+		time.Sleep(time.Duration(waitSeconds) * time.Second)
+	}
+
+	return nil, nil, fmt.Errorf("aliAsyncTaskWait timeout")
+}
+
+func responseAli2OpenAIImage(response *TaskResponse, responseFormat string) *openai.ImageResponse {
+	imageResponse := openai.ImageResponse{
+		Created: helper.GetTimestamp(),
+	}
+
+	for _, data := range response.Output.Results {
+		var b64Json string
+		if responseFormat == "b64_json" {
+			// 读取 data.Url 的图片数据并转存到 b64Json
+			imageData, err := getImageData(data.Url)
+			if err != nil {
+				// 处理获取图片数据失败的情况
+				logger.SysError("getImageData Error getting image data: " + err.Error())
+				continue
+			}
+
+			// 将图片数据转为 Base64 编码的字符串
+			b64Json = Base64Encode(imageData)
+		} else {
+			// 如果 responseFormat 不是 "b64_json"，则直接使用 data.B64Image
+			b64Json = data.B64Image
+		}
+
+		imageResponse.Data = append(imageResponse.Data, openai.ImageData{
+			Url:           data.Url,
+			B64Json:       b64Json,
+			RevisedPrompt: "",
+		})
+	}
+	return &imageResponse
+}
+
+func getImageData(url string) ([]byte, error) {
+	response, err := http.Get(url)
+	if err != nil {
+		return nil, err
+	}
+	defer response.Body.Close()
+
+	imageData, err := io.ReadAll(response.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	return imageData, nil
+}
+
+func Base64Encode(data []byte) string {
+	b64Json := base64.StdEncoding.EncodeToString(data)
+	return b64Json
+}

relay/adaptor/ali/main.go

@@ -7,7 +7,7 @@ import (
 	"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/channel/openai"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/model"
 	"io"
 	"net/http"
@@ -48,6 +48,9 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 			MaxTokens:         request.MaxTokens,
 			Temperature:       request.Temperature,
 			TopP:              request.TopP,
+			TopK:              request.TopK,
+			ResultFormat:      "message",
+			Tools:             request.Tools,
 		},
 	}
 }
@@ -63,6 +66,17 @@ func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) *EmbeddingReque
 	}
 }
 
+func ConvertImageRequest(request model.ImageRequest) *ImageRequest {
+	var imageRequest ImageRequest
+	imageRequest.Input.Prompt = request.Prompt
+	imageRequest.Model = request.Model
+	imageRequest.Parameters.Size = strings.Replace(request.Size, "x", "*", -1)
+	imageRequest.Parameters.N = request.N
+	imageRequest.ResponseFormat = request.ResponseFormat
+
+	return &imageRequest
+}
+
 func EmbeddingHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
 	var aliResponse EmbeddingResponse
 	err := json.NewDecoder(resp.Body).Decode(&aliResponse)
@@ -117,19 +131,11 @@ func embeddingResponseAli2OpenAI(response *EmbeddingResponse) *openai.EmbeddingR
 }
 
 func responseAli2OpenAI(response *ChatResponse) *openai.TextResponse {
-	choice := openai.TextResponseChoice{
-		Index: 0,
-		Message: model.Message{
-			Role:    "assistant",
-			Content: response.Output.Text,
-		},
-		FinishReason: response.Output.FinishReason,
-	}
 	fullTextResponse := openai.TextResponse{
 		Id:      response.RequestId,
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
-		Choices: []openai.TextResponseChoice{choice},
+		Choices: response.Output.Choices,
 		Usage: model.Usage{
 			PromptTokens:     response.Usage.InputTokens,
 			CompletionTokens: response.Usage.OutputTokens,
@@ -140,10 +146,14 @@ func responseAli2OpenAI(response *ChatResponse) *openai.TextResponse {
 }
 
 func streamResponseAli2OpenAI(aliResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
+	if len(aliResponse.Output.Choices) == 0 {
+		return nil
+	}
+	aliChoice := aliResponse.Output.Choices[0]
 	var choice openai.ChatCompletionsStreamResponseChoice
-	choice.Delta.Content = aliResponse.Output.Text
-	if aliResponse.Output.FinishReason != "null" {
-		finishReason := aliResponse.Output.FinishReason
+	choice.Delta = aliChoice.Message
+	if aliChoice.FinishReason != "null" {
+		finishReason := aliChoice.FinishReason
 		choice.FinishReason = &finishReason
 	}
 	response := openai.ChatCompletionsStreamResponse{
@@ -204,6 +214,9 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 				usage.TotalTokens = aliResponse.Usage.InputTokens + aliResponse.Usage.OutputTokens
 			}
 			response := streamResponseAli2OpenAI(&aliResponse)
+			if response == nil {
+				return true
+			}
 			//response.Choices[0].Delta.Content = strings.TrimPrefix(response.Choices[0].Delta.Content, lastResponseText)
 			//lastResponseText = aliResponse.Output.Text
 			jsonResponse, err := json.Marshal(response)
@@ -226,6 +239,7 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 }
 
 func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	ctx := c.Request.Context()
 	var aliResponse ChatResponse
 	responseBody, err := io.ReadAll(resp.Body)
 	if err != nil {
@@ -235,6 +249,7 @@ func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
+	logger.Debugf(ctx, "response body: %s\n", responseBody)
 	err = json.Unmarshal(responseBody, &aliResponse)
 	if err != nil {
 		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil

relay/adaptor/ali/model.go

@@ -0,0 +1,154 @@
+package ali
+
+import (
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+type Message struct {
+	Content string `json:"content"`
+	Role    string `json:"role"`
+}
+
+type Input struct {
+	//Prompt   string       `json:"prompt"`
+	Messages []Message `json:"messages"`
+}
+
+type Parameters struct {
+	TopP              float64      `json:"top_p,omitempty"`
+	TopK              int          `json:"top_k,omitempty"`
+	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"`
+	ResultFormat      string       `json:"result_format,omitempty"`
+	Tools             []model.Tool `json:"tools,omitempty"`
+}
+
+type ChatRequest struct {
+	Model      string     `json:"model"`
+	Input      Input      `json:"input"`
+	Parameters Parameters `json:"parameters,omitempty"`
+}
+
+type ImageRequest struct {
+	Model string `json:"model"`
+	Input struct {
+		Prompt         string `json:"prompt"`
+		NegativePrompt string `json:"negative_prompt,omitempty"`
+	} `json:"input"`
+	Parameters struct {
+		Size  string `json:"size,omitempty"`
+		N     int    `json:"n,omitempty"`
+		Steps string `json:"steps,omitempty"`
+		Scale string `json:"scale,omitempty"`
+	} `json:"parameters,omitempty"`
+	ResponseFormat string `json:"response_format,omitempty"`
+}
+
+type TaskResponse struct {
+	StatusCode int    `json:"status_code,omitempty"`
+	RequestId  string `json:"request_id,omitempty"`
+	Code       string `json:"code,omitempty"`
+	Message    string `json:"message,omitempty"`
+	Output     struct {
+		TaskId     string `json:"task_id,omitempty"`
+		TaskStatus string `json:"task_status,omitempty"`
+		Code       string `json:"code,omitempty"`
+		Message    string `json:"message,omitempty"`
+		Results    []struct {
+			B64Image string `json:"b64_image,omitempty"`
+			Url      string `json:"url,omitempty"`
+			Code     string `json:"code,omitempty"`
+			Message  string `json:"message,omitempty"`
+		} `json:"results,omitempty"`
+		TaskMetrics struct {
+			Total     int `json:"TOTAL,omitempty"`
+			Succeeded int `json:"SUCCEEDED,omitempty"`
+			Failed    int `json:"FAILED,omitempty"`
+		} `json:"task_metrics,omitempty"`
+	} `json:"output,omitempty"`
+	Usage Usage `json:"usage"`
+}
+
+type Header struct {
+	Action       string `json:"action,omitempty"`
+	Streaming    string `json:"streaming,omitempty"`
+	TaskID       string `json:"task_id,omitempty"`
+	Event        string `json:"event,omitempty"`
+	ErrorCode    string `json:"error_code,omitempty"`
+	ErrorMessage string `json:"error_message,omitempty"`
+	Attributes   any    `json:"attributes,omitempty"`
+}
+
+type Payload struct {
+	Model      string `json:"model,omitempty"`
+	Task       string `json:"task,omitempty"`
+	TaskGroup  string `json:"task_group,omitempty"`
+	Function   string `json:"function,omitempty"`
+	Parameters struct {
+		SampleRate int     `json:"sample_rate,omitempty"`
+		Rate       float64 `json:"rate,omitempty"`
+		Format     string  `json:"format,omitempty"`
+	} `json:"parameters,omitempty"`
+	Input struct {
+		Text string `json:"text,omitempty"`
+	} `json:"input,omitempty"`
+	Usage struct {
+		Characters int `json:"characters,omitempty"`
+	} `json:"usage,omitempty"`
+}
+
+type WSSMessage struct {
+	Header  Header  `json:"header,omitempty"`
+	Payload Payload `json:"payload,omitempty"`
+}
+
+type EmbeddingRequest struct {
+	Model string `json:"model"`
+	Input struct {
+		Texts []string `json:"texts"`
+	} `json:"input"`
+	Parameters *struct {
+		TextType string `json:"text_type,omitempty"`
+	} `json:"parameters,omitempty"`
+}
+
+type Embedding struct {
+	Embedding []float64 `json:"embedding"`
+	TextIndex int       `json:"text_index"`
+}
+
+type EmbeddingResponse struct {
+	Output struct {
+		Embeddings []Embedding `json:"embeddings"`
+	} `json:"output"`
+	Usage Usage `json:"usage"`
+	Error
+}
+
+type Error struct {
+	Code      string `json:"code"`
+	Message   string `json:"message"`
+	RequestId string `json:"request_id"`
+}
+
+type Usage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+	TotalTokens  int `json:"total_tokens"`
+}
+
+type Output struct {
+	//Text         string                      `json:"text"`
+	//FinishReason string                      `json:"finish_reason"`
+	Choices []openai.TextResponseChoice `json:"choices"`
+}
+
+type ChatResponse struct {
+	Output Output `json:"output"`
+	Usage  Usage  `json:"usage"`
+	Error
+}

relay/adaptor/anthropic/adaptor.go

@@ -4,9 +4,9 @@ import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/channel"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 )
@@ -14,16 +14,16 @@ import (
 type Adaptor struct {
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
+func (a *Adaptor) Init(meta *meta.Meta) {
 
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	return fmt.Sprintf("%s/v1/messages", meta.BaseURL), nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
 	req.Header.Set("x-api-key", meta.APIKey)
 	anthropicVersion := c.Request.Header.Get("anthropic-version")
 	if anthropicVersion == "" {
@@ -41,11 +41,18 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return ConvertRequest(*request), nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
+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) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+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)
 	} else {

relay/adaptor/anthropic/main.go

@@ -4,16 +4,17 @@ 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"
 	"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/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/model"
-	"io"
-	"net/http"
-	"strings"
 )
 
 func stopReasonClaude2OpenAI(reason *string) string {
@@ -38,6 +39,7 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 		MaxTokens:   textRequest.MaxTokens,
 		Temperature: textRequest.Temperature,
 		TopP:        textRequest.TopP,
+		TopK:        textRequest.TopK,
 		Stream:      textRequest.Stream,
 	}
 	if claudeRequest.MaxTokens == 0 {
@@ -90,7 +92,7 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 }
 
 // https://docs.anthropic.com/claude/reference/messages-streaming
-func streamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
+func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
 	var response *Response
 	var responseText string
 	var stopReason string
@@ -128,7 +130,7 @@ func streamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCo
 	return &openaiResponse, response
 }
 
-func responseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
+func ResponseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
 	var responseText string
 	if len(claudeResponse.Content) > 0 {
 		responseText = claudeResponse.Content[0].Text
@@ -175,10 +177,10 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 			if len(data) < 6 {
 				continue
 			}
-			if !strings.HasPrefix(data, "data: ") {
+			if !strings.HasPrefix(data, "data:") {
 				continue
 			}
-			data = strings.TrimPrefix(data, "data: ")
+			data = strings.TrimPrefix(data, "data:")
 			dataChan <- data
 		}
 		stopChan <- true
@@ -191,14 +193,14 @@ 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 {
 				logger.SysError("error unmarshalling stream response: " + err.Error())
 				return true
 			}
-			response, meta := streamResponseClaude2OpenAI(&claudeResponse)
+			response, meta := StreamResponseClaude2OpenAI(&claudeResponse)
 			if meta != nil {
 				usage.PromptTokens += meta.Usage.InputTokens
 				usage.CompletionTokens += meta.Usage.OutputTokens
@@ -253,7 +255,7 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 			StatusCode: resp.StatusCode,
 		}, nil
 	}
-	fullTextResponse := responseClaude2OpenAI(&claudeResponse)
+	fullTextResponse := ResponseClaude2OpenAI(&claudeResponse)
 	fullTextResponse.Model = modelName
 	usage := model.Usage{
 		PromptTokens:     claudeResponse.Usage.InputTokens,

relay/adaptor/aws/adapter.go

@@ -0,0 +1,82 @@
+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"
+
+	"github.com/gin-gonic/gin"
+	"github.com/pkg/errors"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/anthropic"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+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) {
+	return "", nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	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")
+	}
+
+	claudeReq := anthropic.ConvertRequest(*request)
+	c.Set(ctxkey.RequestModel, request.Model)
+	c.Set(ctxkey.ConvertedRequest, claudeReq)
+	return claudeReq, 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 nil, nil
+}
+
+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, a.awsClient)
+	} else {
+		err, usage = Handler(c, a.awsClient, meta.ActualModelName)
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() (models []string) {
+	for n := range awsModelIDMap {
+		models = append(models, n)
+	}
+	return
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "aws"
+}

relay/adaptor/aws/main.go

@@ -0,0 +1,191 @@
+// Package aws provides the AWS adaptor for the relay service.
+package aws
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"github.com/songquanpeng/one-api/common/ctxkey"
+	"io"
+	"net/http"
+
+	"github.com/aws/aws-sdk-go-v2/aws"
+	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
+	"github.com/aws/aws-sdk-go-v2/service/bedrockruntime/types"
+	"github.com/gin-gonic/gin"
+	"github.com/jinzhu/copier"
+	"github.com/pkg/errors"
+	"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/anthropic"
+	relaymodel "github.com/songquanpeng/one-api/relay/model"
+)
+
+func wrapErr(err error) *relaymodel.ErrorWithStatusCode {
+	return &relaymodel.ErrorWithStatusCode{
+		StatusCode: http.StatusInternalServerError,
+		Error: relaymodel.Error{
+			Message: fmt.Sprintf("%s", err.Error()),
+		},
+	}
+}
+
+// https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html
+var awsModelIDMap = map[string]string{
+	"claude-instant-1.2":       "anthropic.claude-instant-v1",
+	"claude-2.0":               "anthropic.claude-v2",
+	"claude-2.1":               "anthropic.claude-v2:1",
+	"claude-3-sonnet-20240229": "anthropic.claude-3-sonnet-20240229-v1:0",
+	"claude-3-opus-20240229":   "anthropic.claude-3-opus-20240229-v1:0",
+	"claude-3-haiku-20240307":  "anthropic.claude-3-haiku-20240307-v1:0",
+}
+
+func awsModelID(requestModel string) (string, error) {
+	if awsModelID, ok := awsModelIDMap[requestModel]; ok {
+		return awsModelID, nil
+	}
+
+	return "", errors.Errorf("model %s not found", requestModel)
+}
+
+func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
+	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "awsModelID")), nil
+	}
+
+	awsReq := &bedrockruntime.InvokeModelInput{
+		ModelId:     aws.String(awsModelId),
+		Accept:      aws.String("application/json"),
+		ContentType: aws.String("application/json"),
+	}
+
+	claudeReq_, ok := c.Get(ctxkey.ConvertedRequest)
+	if !ok {
+		return wrapErr(errors.New("request not found")), nil
+	}
+	claudeReq := claudeReq_.(*anthropic.Request)
+	awsClaudeReq := &Request{
+		AnthropicVersion: "bedrock-2023-05-31",
+	}
+	if err = copier.Copy(awsClaudeReq, claudeReq); err != nil {
+		return wrapErr(errors.Wrap(err, "copy request")), nil
+	}
+
+	awsReq.Body, err = json.Marshal(awsClaudeReq)
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "marshal request")), nil
+	}
+
+	awsResp, err := awsCli.InvokeModel(c.Request.Context(), awsReq)
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "InvokeModel")), nil
+	}
+
+	claudeResponse := new(anthropic.Response)
+	err = json.Unmarshal(awsResp.Body, claudeResponse)
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "unmarshal response")), nil
+	}
+
+	openaiResp := anthropic.ResponseClaude2OpenAI(claudeResponse)
+	openaiResp.Model = modelName
+	usage := relaymodel.Usage{
+		PromptTokens:     claudeResponse.Usage.InputTokens,
+		CompletionTokens: claudeResponse.Usage.OutputTokens,
+		TotalTokens:      claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens,
+	}
+	openaiResp.Usage = usage
+
+	c.JSON(http.StatusOK, openaiResp)
+	return nil, &usage
+}
+
+func StreamHandler(c *gin.Context, awsCli *bedrockruntime.Client) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
+	createdTime := helper.GetTimestamp()
+	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "awsModelID")), nil
+	}
+
+	awsReq := &bedrockruntime.InvokeModelWithResponseStreamInput{
+		ModelId:     aws.String(awsModelId),
+		Accept:      aws.String("application/json"),
+		ContentType: aws.String("application/json"),
+	}
+
+	claudeReq_, ok := c.Get(ctxkey.ConvertedRequest)
+	if !ok {
+		return wrapErr(errors.New("request not found")), nil
+	}
+	claudeReq := claudeReq_.(*anthropic.Request)
+
+	awsClaudeReq := &Request{
+		AnthropicVersion: "bedrock-2023-05-31",
+	}
+	if err = copier.Copy(awsClaudeReq, claudeReq); err != nil {
+		return wrapErr(errors.Wrap(err, "copy request")), nil
+	}
+	awsReq.Body, err = json.Marshal(awsClaudeReq)
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "marshal request")), nil
+	}
+
+	awsResp, err := awsCli.InvokeModelWithResponseStream(c.Request.Context(), awsReq)
+	if err != nil {
+		return wrapErr(errors.Wrap(err, "InvokeModelWithResponseStream")), nil
+	}
+	stream := awsResp.GetStream()
+	defer stream.Close()
+
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	var usage relaymodel.Usage
+	var id string
+	c.Stream(func(w io.Writer) bool {
+		event, ok := <-stream.Events()
+		if !ok {
+			c.Render(-1, common.CustomEvent{Data: "data: [DONE]"})
+			return false
+		}
+
+		switch v := event.(type) {
+		case *types.ResponseStreamMemberChunk:
+			claudeResp := new(anthropic.StreamResponse)
+			err := json.NewDecoder(bytes.NewReader(v.Value.Bytes)).Decode(claudeResp)
+			if err != nil {
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return false
+			}
+
+			response, meta := anthropic.StreamResponseClaude2OpenAI(claudeResp)
+			if meta != nil {
+				usage.PromptTokens += meta.Usage.InputTokens
+				usage.CompletionTokens += meta.Usage.OutputTokens
+				id = fmt.Sprintf("chatcmpl-%s", meta.Id)
+				return true
+			}
+			if response == nil {
+				return true
+			}
+			response.Id = id
+			response.Model = c.GetString(ctxkey.OriginalModel)
+			response.Created = createdTime
+			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 *types.UnknownUnionMember:
+			fmt.Println("unknown tag:", v.Tag)
+			return false
+		default:
+			fmt.Println("union is nil or unknown type")
+			return false
+		}
+	})
+
+	return nil, &usage
+}

relay/adaptor/aws/model.go

@@ -0,0 +1,17 @@
+package aws
+
+import "github.com/songquanpeng/one-api/relay/adaptor/anthropic"
+
+// Request is the request to AWS Claude
+//
+// https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
+type Request struct {
+	// AnthropicVersion should be "bedrock-2023-05-31"
+	AnthropicVersion string              `json:"anthropic_version"`
+	Messages         []anthropic.Message `json:"messages"`
+	MaxTokens        int                 `json:"max_tokens,omitempty"`
+	Temperature      float64             `json:"temperature,omitempty"`
+	TopP             float64             `json:"top_p,omitempty"`
+	TopK             int                 `json:"top_k,omitempty"`
+	StopSequences    []string            `json:"stop_sequences,omitempty"`
+}

relay/adaptor/baidu/adaptor.go

@@ -3,25 +3,25 @@ package baidu
 import (
 	"errors"
 	"fmt"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/relaymode"
 	"io"
 	"net/http"
 	"strings"
 
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/channel"
-	"github.com/songquanpeng/one-api/relay/constant"
+	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 )
 
 type Adaptor struct {
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
+func (a *Adaptor) Init(meta *meta.Meta) {
 
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/clntwmv7t
 	suffix := "chat/"
 	if strings.HasPrefix(meta.ActualModelName, "Embedding") {
@@ -38,16 +38,34 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 		suffix += "completions_pro"
 	case "ERNIE-Bot-4":
 		suffix += "completions_pro"
-	case "ERNIE-3.5-8K":
-		suffix += "completions"
-	case "ERNIE-Bot-8K":
-		suffix += "ernie_bot_8k"
 	case "ERNIE-Bot":
 		suffix += "completions"
-	case "ERNIE-Speed":
-		suffix += "ernie_speed"
 	case "ERNIE-Bot-turbo":
 		suffix += "eb-instant"
+	case "ERNIE-Speed":
+		suffix += "ernie_speed"
+	case "ERNIE-4.0-8K":
+		suffix += "completions_pro"
+	case "ERNIE-3.5-8K":
+		suffix += "completions"
+	case "ERNIE-3.5-8K-0205":
+		suffix += "ernie-3.5-8k-0205"
+	case "ERNIE-3.5-8K-1222":
+		suffix += "ernie-3.5-8k-1222"
+	case "ERNIE-Bot-8K":
+		suffix += "ernie_bot_8k"
+	case "ERNIE-3.5-4K-0205":
+		suffix += "ernie-3.5-4k-0205"
+	case "ERNIE-Speed-8K":
+		suffix += "ernie_speed"
+	case "ERNIE-Speed-128K":
+		suffix += "ernie-speed-128k"
+	case "ERNIE-Lite-8K-0922":
+		suffix += "eb-instant"
+	case "ERNIE-Lite-8K-0308":
+		suffix += "ernie-lite-8k"
+	case "ERNIE-Tiny-8K":
+		suffix += "ernie-tiny-8k"
 	case "BLOOMZ-7B":
 		suffix += "bloomz_7b1"
 	case "Embedding-V1":
@@ -59,7 +77,7 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 	case "tao-8k":
 		suffix += "tao_8k"
 	default:
-		suffix += meta.ActualModelName
+		suffix += strings.ToLower(meta.ActualModelName)
 	}
 	fullRequestURL := fmt.Sprintf("%s/rpc/2.0/ai_custom/v1/wenxinworkshop/%s", meta.BaseURL, suffix)
 	var accessToken string
@@ -71,8 +89,8 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 	return fullRequestURL, nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+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
 }
@@ -82,7 +100,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		return nil, errors.New("request is nil")
 	}
 	switch relayMode {
-	case constant.RelayModeEmbeddings:
+	case relaymode.Embeddings:
 		baiduEmbeddingRequest := ConvertEmbeddingRequest(*request)
 		return baiduEmbeddingRequest, nil
 	default:
@@ -91,16 +109,23 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
+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) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+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)
 	} else {
 		switch meta.Mode {
-		case constant.RelayModeEmbeddings:
+		case relaymode.Embeddings:
 			err, usage = EmbeddingHandler(c, resp)
 		default:
 			err, usage = Handler(c, resp)

relay/adaptor/baidu/constants.go

@@ -0,0 +1,20 @@
+package baidu
+
+var ModelList = []string{
+	"ERNIE-4.0-8K",
+	"ERNIE-3.5-8K",
+	"ERNIE-3.5-8K-0205",
+	"ERNIE-3.5-8K-1222",
+	"ERNIE-Bot-8K",
+	"ERNIE-3.5-4K-0205",
+	"ERNIE-Speed-8K",
+	"ERNIE-Speed-128K",
+	"ERNIE-Lite-8K-0922",
+	"ERNIE-Lite-8K-0308",
+	"ERNIE-Tiny-8K",
+	"BLOOMZ-7B",
+	"Embedding-V1",
+	"bge-large-zh",
+	"bge-large-en",
+	"tao-8k",
+}

relay/adaptor/baidu/main.go

@@ -8,10 +8,10 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/logger"
-	"github.com/songquanpeng/one-api/relay/channel/openai"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/client"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 	"strings"
@@ -305,7 +305,7 @@ func getBaiduAccessTokenHelper(apiKey string) (*AccessToken, error) {
 	}
 	req.Header.Add("Content-Type", "application/json")
 	req.Header.Add("Accept", "application/json")
-	res, err := util.ImpatientHTTPClient.Do(req)
+	res, err := client.ImpatientHTTPClient.Do(req)
 	if err != nil {
 		return nil, err
 	}

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/adaptor.go

@@ -0,0 +1,64 @@
+package cohere
+
+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{}
+
+// 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) {
+
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	return fmt.Sprintf("%s/v1/chat", 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", "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)
+	} else {
+		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "Cohere"
+}

relay/adaptor/cohere/constant.go

@@ -0,0 +1,14 @@
+package cohere
+
+var ModelList = []string{
+	"command", "command-nightly",
+	"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

@@ -0,0 +1,241 @@
+package cohere
+
+import (
+	"bufio"
+	"bytes"
+	"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"
+	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+var (
+	WebSearchConnector = Connector{ID: "web-search"}
+)
+
+func stopReasonCohere2OpenAI(reason *string) string {
+	if reason == nil {
+		return ""
+	}
+	switch *reason {
+	case "COMPLETE":
+		return "stop"
+	default:
+		return *reason
+	}
+}
+
+func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
+	cohereRequest := Request{
+		Model:            textRequest.Model,
+		Message:          "",
+		MaxTokens:        textRequest.MaxTokens,
+		Temperature:      textRequest.Temperature,
+		P:                textRequest.TopP,
+		K:                textRequest.TopK,
+		Stream:           textRequest.Stream,
+		FrequencyPenalty: textRequest.FrequencyPenalty,
+		PresencePenalty:  textRequest.FrequencyPenalty,
+		Seed:             int(textRequest.Seed),
+	}
+	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)
+		} else {
+			var role string
+			if message.Role == "assistant" {
+				role = "CHATBOT"
+			} else if message.Role == "system" {
+				role = "SYSTEM"
+			} else {
+				role = "USER"
+			}
+			cohereRequest.ChatHistory = append(cohereRequest.ChatHistory, ChatMessage{
+				Role:    role,
+				Message: message.Content.(string),
+			})
+		}
+	}
+	return &cohereRequest
+}
+
+func StreamResponseCohere2OpenAI(cohereResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
+	var response *Response
+	var responseText string
+	var finishReason string
+
+	switch cohereResponse.EventType {
+	case "stream-start":
+		return nil, nil
+	case "text-generation":
+		responseText += cohereResponse.Text
+	case "stream-end":
+		usage := cohereResponse.Response.Meta.Tokens
+		response = &Response{
+			Meta: Meta{
+				Tokens: Usage{
+					InputTokens:  usage.InputTokens,
+					OutputTokens: usage.OutputTokens,
+				},
+			},
+		}
+		finishReason = *cohereResponse.Response.FinishReason
+	default:
+		return nil, nil
+	}
+
+	var choice openai.ChatCompletionsStreamResponseChoice
+	choice.Delta.Content = responseText
+	choice.Delta.Role = "assistant"
+	if finishReason != "" {
+		choice.FinishReason = &finishReason
+	}
+	var openaiResponse openai.ChatCompletionsStreamResponse
+	openaiResponse.Object = "chat.completion.chunk"
+	openaiResponse.Choices = []openai.ChatCompletionsStreamResponseChoice{choice}
+	return &openaiResponse, response
+}
+
+func ResponseCohere2OpenAI(cohereResponse *Response) *openai.TextResponse {
+	choice := openai.TextResponseChoice{
+		Index: 0,
+		Message: model.Message{
+			Role:    "assistant",
+			Content: cohereResponse.Text,
+			Name:    nil,
+		},
+		FinishReason: stopReasonCohere2OpenAI(cohereResponse.FinishReason),
+	}
+	fullTextResponse := openai.TextResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", cohereResponse.ResponseID),
+		Model:   "model",
+		Object:  "chat.completion",
+		Created: helper.GetTimestamp(),
+		Choices: []openai.TextResponseChoice{choice},
+	}
+	return &fullTextResponse
+}
+
+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 := 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()
+			dataChan <- data
+		}
+		stopChan <- true
+	}()
+	common.SetEventStreamHeaders(c)
+	var usage model.Usage
+	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 cohereResponse StreamResponse
+			err := json.Unmarshal([]byte(data), &cohereResponse)
+			if err != nil {
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return true
+			}
+			response, meta := StreamResponseCohere2OpenAI(&cohereResponse)
+			if meta != nil {
+				usage.PromptTokens += meta.Meta.Tokens.InputTokens
+				usage.CompletionTokens += meta.Meta.Tokens.OutputTokens
+				return true
+			}
+			if response == nil {
+				return true
+			}
+			response.Id = fmt.Sprintf("chatcmpl-%d", createdTime)
+			response.Model = c.GetString("original_model")
+			response.Created = createdTime
+			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()
+	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 cohereResponse Response
+	err = json.Unmarshal(responseBody, &cohereResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	if cohereResponse.ResponseID == "" {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: cohereResponse.Message,
+				Type:    cohereResponse.Message,
+				Param:   "",
+				Code:    resp.StatusCode,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+	fullTextResponse := ResponseCohere2OpenAI(&cohereResponse)
+	fullTextResponse.Model = modelName
+	usage := model.Usage{
+		PromptTokens:     cohereResponse.Meta.Tokens.InputTokens,
+		CompletionTokens: cohereResponse.Meta.Tokens.OutputTokens,
+		TotalTokens:      cohereResponse.Meta.Tokens.InputTokens + cohereResponse.Meta.Tokens.OutputTokens,
+	}
+	fullTextResponse.Usage = usage
+	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/cohere/model.go

@@ -0,0 +1,147 @@
+package cohere
+
+type Request struct {
+	Message          string        `json:"message" required:"true"`
+	Model            string        `json:"model,omitempty"`  // 默认值为"command-r"
+	Stream           bool          `json:"stream,omitempty"` // 默认值为false
+	Preamble         string        `json:"preamble,omitempty"`
+	ChatHistory      []ChatMessage `json:"chat_history,omitempty"`
+	ConversationID   string        `json:"conversation_id,omitempty"`
+	PromptTruncation string        `json:"prompt_truncation,omitempty"` // 默认值为"AUTO"
+	Connectors       []Connector   `json:"connectors,omitempty"`
+	Documents        []Document    `json:"documents,omitempty"`
+	Temperature      float64       `json:"temperature,omitempty"` // 默认值为0.3
+	MaxTokens        int           `json:"max_tokens,omitempty"`
+	MaxInputTokens   int           `json:"max_input_tokens,omitempty"`
+	K                int           `json:"k,omitempty"` // 默认值为0
+	P                float64       `json:"p,omitempty"` // 默认值为0.75
+	Seed             int           `json:"seed,omitempty"`
+	StopSequences    []string      `json:"stop_sequences,omitempty"`
+	FrequencyPenalty float64       `json:"frequency_penalty,omitempty"` // 默认值为0.0
+	PresencePenalty  float64       `json:"presence_penalty,omitempty"`  // 默认值为0.0
+	Tools            []Tool        `json:"tools,omitempty"`
+	ToolResults      []ToolResult  `json:"tool_results,omitempty"`
+}
+
+type ChatMessage struct {
+	Role    string `json:"role" required:"true"`
+	Message string `json:"message" required:"true"`
+}
+
+type Tool struct {
+	Name                 string                   `json:"name" required:"true"`
+	Description          string                   `json:"description" required:"true"`
+	ParameterDefinitions map[string]ParameterSpec `json:"parameter_definitions"`
+}
+
+type ParameterSpec struct {
+	Description string `json:"description"`
+	Type        string `json:"type" required:"true"`
+	Required    bool   `json:"required"`
+}
+
+type ToolResult struct {
+	Call    ToolCall                 `json:"call"`
+	Outputs []map[string]interface{} `json:"outputs"`
+}
+
+type ToolCall struct {
+	Name       string                 `json:"name" required:"true"`
+	Parameters map[string]interface{} `json:"parameters" required:"true"`
+}
+
+type StreamResponse struct {
+	IsFinished    bool            `json:"is_finished"`
+	EventType     string          `json:"event_type"`
+	GenerationID  string          `json:"generation_id,omitempty"`
+	SearchQueries []*SearchQuery  `json:"search_queries,omitempty"`
+	SearchResults []*SearchResult `json:"search_results,omitempty"`
+	Documents     []*Document     `json:"documents,omitempty"`
+	Text          string          `json:"text,omitempty"`
+	Citations     []*Citation     `json:"citations,omitempty"`
+	Response      *Response       `json:"response,omitempty"`
+	FinishReason  string          `json:"finish_reason,omitempty"`
+}
+
+type SearchQuery struct {
+	Text         string `json:"text"`
+	GenerationID string `json:"generation_id"`
+}
+
+type SearchResult struct {
+	SearchQuery *SearchQuery `json:"search_query"`
+	DocumentIDs []string     `json:"document_ids"`
+	Connector   *Connector   `json:"connector"`
+}
+
+type Connector struct {
+	ID string `json:"id"`
+}
+
+type Document struct {
+	ID        string `json:"id"`
+	Snippet   string `json:"snippet"`
+	Timestamp string `json:"timestamp"`
+	Title     string `json:"title"`
+	URL       string `json:"url"`
+}
+
+type Citation struct {
+	Start       int      `json:"start"`
+	End         int      `json:"end"`
+	Text        string   `json:"text"`
+	DocumentIDs []string `json:"document_ids"`
+}
+
+type Response struct {
+	ResponseID    string          `json:"response_id"`
+	Text          string          `json:"text"`
+	GenerationID  string          `json:"generation_id"`
+	ChatHistory   []*Message      `json:"chat_history"`
+	FinishReason  *string         `json:"finish_reason"`
+	Meta          Meta            `json:"meta"`
+	Citations     []*Citation     `json:"citations"`
+	Documents     []*Document     `json:"documents"`
+	SearchResults []*SearchResult `json:"search_results"`
+	SearchQueries []*SearchQuery  `json:"search_queries"`
+	Message       string          `json:"message"`
+}
+
+type Message struct {
+	Role    string `json:"role"`
+	Message string `json:"message"`
+}
+
+type Version struct {
+	Version string `json:"version"`
+}
+
+type Units struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}
+
+type ChatEntry struct {
+	Role    string `json:"role"`
+	Message string `json:"message"`
+}
+
+type Meta struct {
+	APIVersion  APIVersion  `json:"api_version"`
+	BilledUnits BilledUnits `json:"billed_units"`
+	Tokens      Usage       `json:"tokens"`
+}
+
+type APIVersion struct {
+	Version string `json:"version"`
+}
+
+type BilledUnits struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}
+
+type Usage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}

relay/adaptor/common.go

@@ -1,15 +1,16 @@
-package channel
+package adaptor
 
 import (
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/util"
+	"github.com/songquanpeng/one-api/relay/client"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"io"
 	"net/http"
 )
 
-func SetupCommonRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) {
+func SetupCommonRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) {
 	req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
 	req.Header.Set("Accept", c.Request.Header.Get("Accept"))
 	if meta.IsStream && c.Request.Header.Get("Accept") == "" {
@@ -17,7 +18,7 @@ func SetupCommonRequestHeader(c *gin.Context, req *http.Request, meta *util.Rela
 	}
 }
 
-func DoRequestHelper(a Adaptor, c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
+func DoRequestHelper(a Adaptor, c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error) {
 	fullRequestURL, err := a.GetRequestURL(meta)
 	if err != nil {
 		return nil, fmt.Errorf("get request url failed: %w", err)
@@ -38,7 +39,7 @@ func DoRequestHelper(a Adaptor, c *gin.Context, meta *util.RelayMeta, requestBod
 }
 
 func DoRequest(c *gin.Context, req *http.Request) (*http.Response, error) {
-	resp, err := util.HTTPClient.Do(req)
+	resp, err := client.HTTPClient.Do(req)
 	if err != nil {
 		return nil, err
 	}

relay/adaptor/coze/adaptor.go

@@ -0,0 +1,75 @@
+package coze
+
+import (
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"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 {
+	meta *meta.Meta
+}
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.meta = meta
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	return fmt.Sprintf("%s/open_api/v2/chat", 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", "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")
+	}
+	request.User = a.meta.Config.UserID
+	return ConvertRequest(*request), 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) {
+	var responseText *string
+	if meta.IsStream {
+		err, responseText = StreamHandler(c, resp)
+	} else {
+		err, responseText = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	}
+	if responseText != nil {
+		usage = openai.ResponseText2Usage(*responseText, meta.ActualModelName, meta.PromptTokens)
+	} else {
+		usage = &model.Usage{}
+	}
+	usage.PromptTokens = meta.PromptTokens
+	usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "coze"
+}

relay/adaptor/coze/constant/contenttype/define.go

@@ -0,0 +1,5 @@
+package contenttype
+
+const (
+	Text = "text"
+)

relay/adaptor/coze/constant/event/define.go

@@ -0,0 +1,7 @@
+package event
+
+const (
+	Message = "message"
+	Done    = "done"
+	Error   = "error"
+)

relay/adaptor/coze/constant/messagetype/define.go

@@ -0,0 +1,6 @@
+package messagetype
+
+const (
+	Answer   = "answer"
+	FollowUp = "follow_up"
+)

relay/adaptor/coze/constants.go

@@ -0,0 +1,3 @@
+package coze
+
+var ModelList = []string{}

relay/adaptor/coze/helper.go

@@ -0,0 +1,10 @@
+package coze
+
+import "github.com/songquanpeng/one-api/relay/adaptor/coze/constant/event"
+
+func event2StopReason(e *string) string {
+	if e == nil || *e == event.Message {
+		return ""
+	}
+	return "stop"
+}

relay/adaptor/coze/main.go

@@ -0,0 +1,215 @@
+package coze
+
+import (
+	"bufio"
+	"encoding/json"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/conv"
+	"github.com/songquanpeng/one-api/common/helper"
+	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/relay/adaptor/coze/constant/messagetype"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
+	"strings"
+)
+
+// https://www.coze.com/open
+
+func stopReasonCoze2OpenAI(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
+	}
+}
+
+func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
+	cozeRequest := Request{
+		Stream: textRequest.Stream,
+		User:   textRequest.User,
+		BotId:  strings.TrimPrefix(textRequest.Model, "bot-"),
+	}
+	for i, message := range textRequest.Messages {
+		if i == len(textRequest.Messages)-1 {
+			cozeRequest.Query = message.StringContent()
+			continue
+		}
+		cozeMessage := Message{
+			Role:    message.Role,
+			Content: message.StringContent(),
+		}
+		cozeRequest.ChatHistory = append(cozeRequest.ChatHistory, cozeMessage)
+	}
+	return &cozeRequest
+}
+
+func StreamResponseCoze2OpenAI(cozeResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
+	var response *Response
+	var stopReason string
+	var choice openai.ChatCompletionsStreamResponseChoice
+
+	if cozeResponse.Message != nil {
+		if cozeResponse.Message.Type != messagetype.Answer {
+			return nil, nil
+		}
+		choice.Delta.Content = cozeResponse.Message.Content
+	}
+	choice.Delta.Role = "assistant"
+	finishReason := stopReasonCoze2OpenAI(&stopReason)
+	if finishReason != "null" {
+		choice.FinishReason = &finishReason
+	}
+	var openaiResponse openai.ChatCompletionsStreamResponse
+	openaiResponse.Object = "chat.completion.chunk"
+	openaiResponse.Choices = []openai.ChatCompletionsStreamResponseChoice{choice}
+	openaiResponse.Id = cozeResponse.ConversationId
+	return &openaiResponse, response
+}
+
+func ResponseCoze2OpenAI(cozeResponse *Response) *openai.TextResponse {
+	var responseText string
+	for _, message := range cozeResponse.Messages {
+		if message.Type == messagetype.Answer {
+			responseText = message.Content
+			break
+		}
+	}
+	choice := openai.TextResponseChoice{
+		Index: 0,
+		Message: model.Message{
+			Role:    "assistant",
+			Content: responseText,
+			Name:    nil,
+		},
+		FinishReason: "stop",
+	}
+	fullTextResponse := openai.TextResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", cozeResponse.ConversationId),
+		Model:   "coze-bot",
+		Object:  "chat.completion",
+		Created: helper.GetTimestamp(),
+		Choices: []openai.TextResponseChoice{choice},
+	}
+	return &fullTextResponse
+}
+
+func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *string) {
+	var responseText string
+	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), "\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) < 5 {
+				continue
+			}
+			if !strings.HasPrefix(data, "data:") {
+				continue
+			}
+			data = strings.TrimPrefix(data, "data:")
+			dataChan <- data
+		}
+		stopChan <- true
+	}()
+	common.SetEventStreamHeaders(c)
+	var modelName 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 cozeResponse StreamResponse
+			err := json.Unmarshal([]byte(data), &cozeResponse)
+			if err != nil {
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return true
+			}
+			response, _ := StreamResponseCoze2OpenAI(&cozeResponse)
+			if response == nil {
+				return true
+			}
+			for _, choice := range response.Choices {
+				responseText += conv.AsString(choice.Delta.Content)
+			}
+			response.Model = modelName
+			response.Created = createdTime
+			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()
+	return nil, &responseText
+}
+
+func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *string) {
+	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 cozeResponse Response
+	err = json.Unmarshal(responseBody, &cozeResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	if cozeResponse.Code != 0 {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: cozeResponse.Msg,
+				Code:    cozeResponse.Code,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+	fullTextResponse := ResponseCoze2OpenAI(&cozeResponse)
+	fullTextResponse.Model = modelName
+	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)
+	var responseText string
+	if len(fullTextResponse.Choices) > 0 {
+		responseText = fullTextResponse.Choices[0].Message.StringContent()
+	}
+	return nil, &responseText
+}

relay/adaptor/coze/model.go

@@ -0,0 +1,38 @@
+package coze
+
+type Message struct {
+	Role        string `json:"role"`
+	Type        string `json:"type"`
+	Content     string `json:"content"`
+	ContentType string `json:"content_type"`
+}
+
+type ErrorInformation struct {
+	Code int    `json:"code"`
+	Msg  string `json:"msg"`
+}
+
+type Request struct {
+	ConversationId string    `json:"conversation_id,omitempty"`
+	BotId          string    `json:"bot_id"`
+	User           string    `json:"user"`
+	Query          string    `json:"query"`
+	ChatHistory    []Message `json:"chat_history,omitempty"`
+	Stream         bool      `json:"stream"`
+}
+
+type Response struct {
+	ConversationId string    `json:"conversation_id,omitempty"`
+	Messages       []Message `json:"messages,omitempty"`
+	Code           int       `json:"code,omitempty"`
+	Msg            string    `json:"msg,omitempty"`
+}
+
+type StreamResponse struct {
+	Event            string            `json:"event,omitempty"`
+	Message          *Message          `json:"message,omitempty"`
+	IsFinish         bool              `json:"is_finish,omitempty"`
+	Index            int               `json:"index,omitempty"`
+	ConversationId   string            `json:"conversation_id,omitempty"`
+	ErrorInformation *ErrorInformation `json:"error_information,omitempty"`
+}

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

@@ -3,33 +3,35 @@ package gemini
 import (
 	"errors"
 	"fmt"
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common/helper"
-	channelhelper "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"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/helper"
+	channelhelper "github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
 )
 
 type Adaptor struct {
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
+func (a *Adaptor) Init(meta *meta.Meta) {
 
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	version := helper.AssignOrDefault(meta.APIVersion, "v1")
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	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
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
 	channelhelper.SetupCommonRequestHeader(c, req, meta)
 	req.Header.Set("x-goog-api-key", meta.APIKey)
 	return nil
@@ -42,11 +44,18 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return ConvertRequest(*request), nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
+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 channelhelper.DoRequestHelper(a, c, meta, requestBody)
 }
 
-func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
 		var responseText string
 		err, responseText = StreamHandler(c, resp)

relay/adaptor/gemini/constants.go

@@ -3,6 +3,6 @@ package gemini
 // https://ai.google.dev/models/gemini
 
 var ModelList = []string{
-	"gemini-pro", "gemini-1.0-pro-001",
+	"gemini-pro", "gemini-1.0-pro-001", "gemini-1.5-pro",
 	"gemini-pro-vision", "gemini-1.0-pro-vision-001",
 }

relay/adaptor/gemini/main.go

@@ -4,17 +4,19 @@ import (
 	"bufio"
 	"encoding/json"
 	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
 	"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/image"
 	"github.com/songquanpeng/one-api/common/logger"
-	"github.com/songquanpeng/one-api/relay/channel/openai"
+	"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/model"
-	"io"
-	"net/http"
-	"strings"
 
 	"github.com/gin-gonic/gin"
 )
@@ -53,7 +55,17 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 			MaxOutputTokens: textRequest.MaxTokens,
 		},
 	}
-	if textRequest.Functions != nil {
+	if textRequest.Tools != nil {
+		functions := make([]model.Function, 0, len(textRequest.Tools))
+		for _, tool := range textRequest.Tools {
+			functions = append(functions, tool.Function)
+		}
+		geminiRequest.Tools = []ChatTools{
+			{
+				FunctionDeclarations: functions,
+			},
+		}
+	} else if textRequest.Functions != nil {
 		geminiRequest.Tools = []ChatTools{
 			{
 				FunctionDeclarations: textRequest.Functions,
@@ -153,9 +165,33 @@ type ChatPromptFeedback struct {
 	SafetyRatings []ChatSafetyRating `json:"safetyRatings"`
 }
 
+func getToolCalls(candidate *ChatCandidate) []model.Tool {
+	var toolCalls []model.Tool
+
+	item := candidate.Content.Parts[0]
+	if item.FunctionCall == nil {
+		return toolCalls
+	}
+	argsBytes, err := json.Marshal(item.FunctionCall.Arguments)
+	if err != nil {
+		logger.FatalLog("getToolCalls failed: " + err.Error())
+		return toolCalls
+	}
+	toolCall := model.Tool{
+		Id:   fmt.Sprintf("call_%s", random.GetUUID()),
+		Type: "function",
+		Function: model.Function{
+			Arguments: string(argsBytes),
+			Name:      item.FunctionCall.FunctionName,
+		},
+	}
+	toolCalls = append(toolCalls, toolCall)
+	return toolCalls
+}
+
 func responseGeminiChat2OpenAI(response *ChatResponse) *openai.TextResponse {
 	fullTextResponse := openai.TextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: make([]openai.TextResponseChoice, 0, len(response.Candidates)),
@@ -164,13 +200,19 @@ func responseGeminiChat2OpenAI(response *ChatResponse) *openai.TextResponse {
 		choice := openai.TextResponseChoice{
 			Index: i,
 			Message: model.Message{
-				Role:    "assistant",
-				Content: "",
+				Role: "assistant",
 			},
 			FinishReason: constant.StopFinishReason,
 		}
 		if len(candidate.Content.Parts) > 0 {
-			choice.Message.Content = candidate.Content.Parts[0].Text
+			if candidate.Content.Parts[0].FunctionCall != nil {
+				choice.Message.ToolCalls = getToolCalls(&candidate)
+			} else {
+				choice.Message.Content = candidate.Content.Parts[0].Text
+			}
+		} else {
+			choice.Message.Content = ""
+			choice.FinishReason = candidate.FinishReason
 		}
 		fullTextResponse.Choices = append(fullTextResponse.Choices, choice)
 	}
@@ -190,8 +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 {
@@ -205,14 +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
 		}
@@ -222,23 +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
-			data = fmt.Sprintf("{\"content\": \"%s\"}", data)
-			type dummyStruct struct {
-				Content string `json:"content"`
+			var geminiResponse ChatResponse
+			err := json.Unmarshal([]byte(data), &geminiResponse)
+			if err != nil {
+				logger.SysError("error unmarshalling stream response: " + err.Error())
+				return true
 			}
-			var dummy dummyStruct
-			err := json.Unmarshal([]byte(data), &dummy)
-			responseText += dummy.Content
-			var choice openai.ChatCompletionsStreamResponseChoice
-			choice.Delta.Content = dummy.Content
-			response := openai.ChatCompletionsStreamResponse{
-				Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
-				Object:  "chat.completion.chunk",
-				Created: helper.GetTimestamp(),
-				Model:   "gemini-pro",
-				Choices: []openai.ChatCompletionsStreamResponseChoice{choice},
+			response := streamResponseGeminiChat2OpenAI(&geminiResponse)
+			if response == nil {
+				return true
 			}
+			responseText += response.Choices[0].Delta.StringContent()
 			jsonResponse, err := json.Marshal(response)
 			if err != nil {
 				logger.SysError("error marshalling stream response: " + err.Error())

relay/adaptor/gemini/model.go

@@ -12,9 +12,15 @@ type InlineData struct {
 	Data     string `json:"data"`
 }
 
+type FunctionCall struct {
+	FunctionName string `json:"name"`
+	Arguments    any    `json:"args"`
+}
+
 type Part struct {
-	Text       string      `json:"text,omitempty"`
-	InlineData *InlineData `json:"inlineData,omitempty"`
+	Text         string        `json:"text,omitempty"`
+	InlineData   *InlineData   `json:"inlineData,omitempty"`
+	FunctionCall *FunctionCall `json:"functionCall,omitempty"`
 }
 
 type ChatContent struct {
@@ -28,7 +34,7 @@ type ChatSafetySettings struct {
 }
 
 type ChatTools struct {
-	FunctionDeclarations any `json:"functionDeclarations,omitempty"`
+	FunctionDeclarations any `json:"function_declarations,omitempty"`
 }
 
 type ChatGenerationConfig struct {

relay/adaptor/groq/constants.go

@@ -7,4 +7,6 @@ var ModelList = []string{
 	"llama2-7b-2048",
 	"llama2-70b-4096",
 	"mixtral-8x7b-32768",
+	"llama3-8b-8192",
+	"llama3-70b-8192",
 }

relay/adaptor/interface.go

@@ -0,0 +1,21 @@
+package adaptor
+
+import (
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
+)
+
+type Adaptor interface {
+	Init(meta *meta.Meta)
+	GetRequestURL(meta *meta.Meta) (string, error)
+	SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error
+	ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error)
+	ConvertImageRequest(request *model.ImageRequest) (any, error)
+	DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error)
+	DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode)
+	GetModelList() []string
+	GetChannelName() string
+}

relay/adaptor/minimax/constants.go

@@ -0,0 +1,11 @@
+package minimax
+
+// https://www.minimaxi.com/document/guides/chat-model/V2?id=65e0736ab2845de20908e2dd
+
+var ModelList = []string{
+	"abab6.5-chat",
+	"abab6.5s-chat",
+	"abab6-chat",
+	"abab5.5-chat",
+	"abab5.5s-chat",
+}

relay/adaptor/minimax/main.go

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

relay/adaptor/ollama/adaptor.go

@@ -0,0 +1,82 @@
+package ollama
+
+import (
+	"errors"
+	"fmt"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/relaymode"
+	"io"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+type Adaptor struct {
+}
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	// https://github.com/ollama/ollama/blob/main/docs/api.md
+	fullRequestURL := fmt.Sprintf("%s/api/chat", meta.BaseURL)
+	if meta.Mode == relaymode.Embeddings {
+		fullRequestURL = fmt.Sprintf("%s/api/embeddings", meta.BaseURL)
+	}
+	return fullRequestURL, 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")
+	}
+	switch relayMode {
+	case relaymode.Embeddings:
+		ollamaEmbeddingRequest := ConvertEmbeddingRequest(*request)
+		return ollamaEmbeddingRequest, nil
+	default:
+		return ConvertRequest(*request), 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, usage = StreamHandler(c, resp)
+	} else {
+		switch meta.Mode {
+		case relaymode.Embeddings:
+			err, usage = EmbeddingHandler(c, resp)
+		default:
+			err, usage = Handler(c, resp)
+		}
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "ollama"
+}

relay/adaptor/ollama/constants.go

@@ -0,0 +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

@@ -5,16 +5,19 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"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/channel/openai"
-	"github.com/songquanpeng/one-api/relay/constant"
-	"github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/common/random"
 	"io"
 	"net/http"
 	"strings"
+
+	"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"
+	"github.com/songquanpeng/one-api/relay/model"
 )
 
 func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
@@ -30,9 +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
@@ -50,7 +66,8 @@ func responseOllama2OpenAI(response *ChatResponse) *openai.TextResponse {
 		choice.FinishReason = "stop"
 	}
 	fullTextResponse := openai.TextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
+		Model:   response.Model,
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -71,7 +88,7 @@ func streamResponseOllama2OpenAI(ollamaResponse *ChatResponse) *openai.ChatCompl
 		choice.FinishReason = &constant.StopFinishReason
 	}
 	response := openai.ChatCompletionsStreamResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   ollamaResponse.Model,
@@ -139,6 +156,64 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 	return nil, &usage
 }
 
+func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) *EmbeddingRequest {
+	return &EmbeddingRequest{
+		Model:  request.Model,
+		Prompt: strings.Join(request.ParseInput(), " "),
+	}
+}
+
+func EmbeddingHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	var ollamaResponse EmbeddingResponse
+	err := json.NewDecoder(resp.Body).Decode(&ollamaResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	err = resp.Body.Close()
+	if err != nil {
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	if ollamaResponse.Error != "" {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: ollamaResponse.Error,
+				Type:    "ollama_error",
+				Param:   "",
+				Code:    "ollama_error",
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+
+	fullTextResponse := embeddingResponseOllama2OpenAI(&ollamaResponse)
+	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, &fullTextResponse.Usage
+}
+
+func embeddingResponseOllama2OpenAI(response *EmbeddingResponse) *openai.EmbeddingResponse {
+	openAIEmbeddingResponse := openai.EmbeddingResponse{
+		Object: "list",
+		Data:   make([]openai.EmbeddingResponseItem, 0, 1),
+		Model:  "text-embedding-v1",
+		Usage:  model.Usage{TotalTokens: 0},
+	}
+
+	openAIEmbeddingResponse.Data = append(openAIEmbeddingResponse.Data, openai.EmbeddingResponseItem{
+		Object:    `embedding`,
+		Index:     0,
+		Embedding: response.Embedding,
+	})
+	return &openAIEmbeddingResponse
+}
+
 func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
 	ctx := context.TODO()
 	var ollamaResponse ChatResponse

relay/adaptor/ollama/model.go

@@ -35,3 +35,13 @@ type ChatResponse struct {
 	EvalDuration    int     `json:"eval_duration,omitempty"`
 	Error           string  `json:"error,omitempty"`
 }
+
+type EmbeddingRequest struct {
+	Model  string `json:"model"`
+	Prompt string `json:"prompt"`
+}
+
+type EmbeddingResponse struct {
+	Error     string    `json:"error,omitempty"`
+	Embedding []float64 `json:"embedding,omitempty"`
+}

relay/adaptor/openai/adaptor.go

@@ -0,0 +1,115 @@
+package openai
+
+import (
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/minimax"
+	"github.com/songquanpeng/one-api/relay/channeltype"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/relay/relaymode"
+	"io"
+	"net/http"
+	"strings"
+)
+
+type Adaptor struct {
+	ChannelType int
+}
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.ChannelType = meta.ChannelType
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	switch meta.ChannelType {
+	case channeltype.Azure:
+		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.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.Config.APIVersion)
+		task := strings.TrimPrefix(requestURL, "/v1/")
+		model_ := meta.ActualModelName
+		model_ = strings.Replace(model_, ".", "", -1)
+		//https://github.com/songquanpeng/one-api/issues/1191
+		// {your endpoint}/openai/deployments/{your azure_model}/chat/completions?api-version={api_version}
+		requestURL = fmt.Sprintf("/openai/deployments/%s/%s", model_, task)
+		return GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType), nil
+	case channeltype.Minimax:
+		return minimax.GetRequestURL(meta)
+	default:
+		return GetFullRequestURL(meta.BaseURL, meta.RequestURLPath, meta.ChannelType), nil
+	}
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
+	if meta.ChannelType == channeltype.Azure {
+		req.Header.Set("api-key", meta.APIKey)
+		return nil
+	}
+	req.Header.Set("Authorization", "Bearer "+meta.APIKey)
+	if meta.ChannelType == channeltype.OpenRouter {
+		req.Header.Set("HTTP-Referer", "https://github.com/songquanpeng/one-api")
+		req.Header.Set("X-Title", "One API")
+	}
+	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 request, 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 {
+		var responseText string
+		err, responseText, usage = StreamHandler(c, resp, meta.Mode)
+		if usage == nil || usage.TotalTokens == 0 {
+			usage = ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
+		}
+		if usage.TotalTokens != 0 && usage.PromptTokens == 0 { // some channels don't return prompt tokens & completion tokens
+			usage.PromptTokens = meta.PromptTokens
+			usage.CompletionTokens = usage.TotalTokens - meta.PromptTokens
+		}
+	} else {
+		switch meta.Mode {
+		case relaymode.ImagesGenerations:
+			err, _ = ImageHandler(c, resp)
+		default:
+			err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+		}
+	}
+	return
+}
+
+func (a *Adaptor) GetModelList() []string {
+	_, modelList := GetCompatibleChannelMeta(a.ChannelType)
+	return modelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	channelName, _ := GetCompatibleChannelMeta(a.ChannelType)
+	return channelName
+}

relay/adaptor/openai/compatible.go

@@ -0,0 +1,58 @@
+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"
+	"github.com/songquanpeng/one-api/relay/adaptor/mistral"
+	"github.com/songquanpeng/one-api/relay/adaptor/moonshot"
+	"github.com/songquanpeng/one-api/relay/adaptor/stepfun"
+	"github.com/songquanpeng/one-api/relay/adaptor/togetherai"
+	"github.com/songquanpeng/one-api/relay/channeltype"
+)
+
+var CompatibleChannels = []int{
+	channeltype.Azure,
+	channeltype.AI360,
+	channeltype.Moonshot,
+	channeltype.Baichuan,
+	channeltype.Minimax,
+	channeltype.Mistral,
+	channeltype.Groq,
+	channeltype.LingYiWanWu,
+	channeltype.StepFun,
+	channeltype.DeepSeek,
+	channeltype.TogetherAI,
+}
+
+func GetCompatibleChannelMeta(channelType int) (string, []string) {
+	switch channelType {
+	case channeltype.Azure:
+		return "azure", ModelList
+	case channeltype.AI360:
+		return "360", ai360.ModelList
+	case channeltype.Moonshot:
+		return "moonshot", moonshot.ModelList
+	case channeltype.Baichuan:
+		return "baichuan", baichuan.ModelList
+	case channeltype.Minimax:
+		return "minimax", minimax.ModelList
+	case channeltype.Mistral:
+		return "mistralai", mistral.ModelList
+	case channeltype.Groq:
+		return "groq", groq.ModelList
+	case channeltype.LingYiWanWu:
+		return "lingyiwanwu", lingyiwanwu.ModelList
+	case channeltype.StepFun:
+		return "stepfun", stepfun.ModelList
+	case channeltype.DeepSeek:
+		return "deepseek", deepseek.ModelList
+	case channeltype.TogetherAI:
+		return "together.ai", togetherai.ModelList
+	default:
+		return "openai", ModelList
+	}
+}

relay/adaptor/openai/constants.go

@@ -6,7 +6,7 @@ var ModelList = []string{
 	"gpt-3.5-turbo-instruct",
 	"gpt-4", "gpt-4-0314", "gpt-4-0613", "gpt-4-1106-preview", "gpt-4-0125-preview",
 	"gpt-4-32k", "gpt-4-32k-0314", "gpt-4-32k-0613",
-	"gpt-4-turbo-preview",
+	"gpt-4-turbo-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09",
 	"gpt-4-vision-preview",
 	"text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large",
 	"text-curie-001", "text-babbage-001", "text-ada-001", "text-davinci-002", "text-davinci-003",

relay/adaptor/openai/helper.go

@@ -0,0 +1,30 @@
+package openai
+
+import (
+	"fmt"
+	"github.com/songquanpeng/one-api/relay/channeltype"
+	"github.com/songquanpeng/one-api/relay/model"
+	"strings"
+)
+
+func ResponseText2Usage(responseText string, modeName string, promptTokens int) *model.Usage {
+	usage := &model.Usage{}
+	usage.PromptTokens = promptTokens
+	usage.CompletionTokens = CountTokenText(responseText, modeName)
+	usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+	return usage
+}
+
+func GetFullRequestURL(baseURL string, requestURL string, channelType int) string {
+	fullRequestURL := fmt.Sprintf("%s%s", baseURL, requestURL)
+
+	if strings.HasPrefix(baseURL, "https://gateway.ai.cloudflare.com") {
+		switch channelType {
+		case channeltype.OpenAI:
+			fullRequestURL = fmt.Sprintf("%s%s", baseURL, strings.TrimPrefix(requestURL, "/v1"))
+		case channeltype.Azure:
+			fullRequestURL = fmt.Sprintf("%s%s", baseURL, strings.TrimPrefix(requestURL, "/openai/deployments"))
+		}
+	}
+	return fullRequestURL
+}

relay/adaptor/openai/image.go

@@ -0,0 +1,44 @@
+package openai
+
+import (
+	"bytes"
+	"encoding/json"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/model"
+	"io"
+	"net/http"
+)
+
+func ImageHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	var imageResponse ImageResponse
+	responseBody, err := io.ReadAll(resp.Body)
+
+	if err != nil {
+		return ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = json.Unmarshal(responseBody, &imageResponse)
+	if err != nil {
+		return ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
+
+	for k, v := range resp.Header {
+		c.Writer.Header().Set(k, v[0])
+	}
+	c.Writer.WriteHeader(resp.StatusCode)
+
+	_, err = io.Copy(c.Writer, resp.Body)
+	if err != nil {
+		return ErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+	return nil, nil
+}

relay/adaptor/openai/main.go

@@ -6,14 +6,21 @@ import (
 	"encoding/json"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/conv"
 	"github.com/songquanpeng/one-api/common/logger"
-	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/relay/relaymode"
 	"io"
 	"net/http"
 	"strings"
 )
 
+const (
+	dataPrefix       = "data: "
+	done             = "[DONE]"
+	dataPrefixLength = len(dataPrefix)
+)
+
 func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
 	responseText := ""
 	scanner := bufio.NewScanner(resp.Body)
@@ -35,39 +42,46 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 	go func() {
 		for scanner.Scan() {
 			data := scanner.Text()
-			if len(data) < 6 { // ignore blank line or wrong format
+			if len(data) < dataPrefixLength { // ignore blank line or wrong format
 				continue
 			}
-			if data[:6] != "data: " && data[:6] != "[DONE]" {
+			if data[:dataPrefixLength] != dataPrefix && data[:dataPrefixLength] != done {
 				continue
 			}
-			dataChan <- data
-			data = data[6:]
-			if !strings.HasPrefix(data, "[DONE]") {
-				switch relayMode {
-				case constant.RelayModeChatCompletions:
-					var streamResponse ChatCompletionsStreamResponse
-					err := json.Unmarshal([]byte(data), &streamResponse)
-					if err != nil {
-						logger.SysError("error unmarshalling stream response: " + err.Error())
-						continue // just ignore the error
-					}
-					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)
-					if err != nil {
-						logger.SysError("error unmarshalling stream response: " + err.Error())
-						continue
-					}
-					for _, choice := range streamResponse.Choices {
-						responseText += choice.Text
-					}
+			if strings.HasPrefix(data[dataPrefixLength:], done) {
+				dataChan <- data
+				continue
+			}
+			switch relayMode {
+			case relaymode.ChatCompletions:
+				var streamResponse ChatCompletionsStreamResponse
+				err := json.Unmarshal([]byte(data[dataPrefixLength:]), &streamResponse)
+				if err != nil {
+					logger.SysError("error unmarshalling stream response: " + err.Error())
+					dataChan <- data // if error happened, pass the data to client
+					continue         // just ignore the error
+				}
+				if len(streamResponse.Choices) == 0 {
+					// but for empty choice, we should not pass it to client, this is for azure
+					continue // just ignore empty choice
+				}
+				dataChan <- data
+				for _, choice := range streamResponse.Choices {
+					responseText += conv.AsString(choice.Delta.Content)
+				}
+				if streamResponse.Usage != nil {
+					usage = streamResponse.Usage
+				}
+			case relaymode.Completions:
+				dataChan <- data
+				var streamResponse CompletionsStreamResponse
+				err := json.Unmarshal([]byte(data[dataPrefixLength:]), &streamResponse)
+				if err != nil {
+					logger.SysError("error unmarshalling stream response: " + err.Error())
+					continue
+				}
+				for _, choice := range streamResponse.Choices {
+					responseText += choice.Text
 				}
 			}
 		}

relay/adaptor/openai/model.go

@@ -110,20 +110,22 @@ type EmbeddingResponse struct {
 	model.Usage `json:"usage"`
 }
 
+type ImageData struct {
+	Url           string `json:"url,omitempty"`
+	B64Json       string `json:"b64_json,omitempty"`
+	RevisedPrompt string `json:"revised_prompt,omitempty"`
+}
+
 type ImageResponse struct {
-	Created int `json:"created"`
-	Data    []struct {
-		Url string `json:"url"`
-	}
+	Created int64       `json:"created"`
+	Data    []ImageData `json:"data"`
+	//model.Usage `json:"usage"`
 }
 
 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"`
+	Index        int           `json:"index"`
+	Delta        model.Message `json:"delta"`
+	FinishReason *string       `json:"finish_reason,omitempty"`
 }
 
 type ChatCompletionsStreamResponse struct {
@@ -132,7 +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

@@ -4,10 +4,10 @@ import (
 	"errors"
 	"fmt"
 	"github.com/pkoukk/tiktoken-go"
-	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
+	billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/model"
 	"math"
 	"strings"
@@ -28,7 +28,7 @@ func InitTokenEncoders() {
 	if err != nil {
 		logger.FatalLog(fmt.Sprintf("failed to get gpt-4 token encoder: %s", err.Error()))
 	}
-	for model := range common.ModelRatio {
+	for model := range billingratio.ModelRatio {
 		if strings.HasPrefix(model, "gpt-3.5") {
 			tokenEncoderMap[model] = gpt35TokenEncoder
 		} else if strings.HasPrefix(model, "gpt-4") {
@@ -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/palm/adaptor.go

@@ -4,10 +4,10 @@ import (
 	"errors"
 	"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/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 )
@@ -15,16 +15,16 @@ import (
 type Adaptor struct {
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
+func (a *Adaptor) Init(meta *meta.Meta) {
 
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	return fmt.Sprintf("%s/v1beta2/models/chat-bison-001:generateMessage", meta.BaseURL), nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
 	req.Header.Set("x-goog-api-key", meta.APIKey)
 	return nil
 }
@@ -36,11 +36,18 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return ConvertRequest(*request), nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
+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) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+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 {
 		var responseText string
 		err, responseText = StreamHandler(c, resp)

relay/adaptor/palm/palm.go

@@ -7,7 +7,8 @@ import (
 	"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/channel/openai"
+	"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/model"
 	"io"
@@ -74,7 +75,7 @@ func streamResponsePaLM2OpenAI(palmResponse *ChatResponse) *openai.ChatCompletio
 
 func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, string) {
 	responseText := ""
-	responseId := fmt.Sprintf("chatcmpl-%s", helper.GetUUID())
+	responseId := fmt.Sprintf("chatcmpl-%s", random.GetUUID())
 	createdTime := helper.GetTimestamp()
 	dataChan := make(chan string)
 	stopChan := make(chan bool)

relay/adaptor/stepfun/constants.go

@@ -0,0 +1,7 @@
+package stepfun
+
+var ModelList = []string{
+	"step-1-32k",
+	"step-1v-32k",
+	"step-1-200k",
+}

relay/adaptor/tencent/adaptor.go

@@ -4,10 +4,10 @@ import (
 	"errors"
 	"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/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 	"strings"
@@ -19,16 +19,16 @@ type Adaptor struct {
 	Sign string
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
+func (a *Adaptor) Init(meta *meta.Meta) {
 
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	return fmt.Sprintf("%s/hyllm/v1/chat/completions", meta.BaseURL), nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
 	req.Header.Set("Authorization", a.Sign)
 	req.Header.Set("X-TC-Action", meta.ActualModelName)
 	return nil
@@ -52,11 +52,18 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return tencentRequest, nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
+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) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+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 {
 		var responseText string
 		err, responseText = StreamHandler(c, resp)

relay/adaptor/tencent/main.go

@@ -10,9 +10,11 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/conv"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/logger"
-	"github.com/songquanpeng/one-api/relay/channel/openai"
+	"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/model"
 	"io"
@@ -40,7 +42,7 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 	return &ChatRequest{
 		Timestamp:   helper.GetTimestamp(),
 		Expired:     helper.GetTimestamp() + 24*60*60,
-		QueryID:     helper.GetUUID(),
+		QueryID:     random.GetUUID(),
 		Temperature: request.Temperature,
 		TopP:        request.TopP,
 		Stream:      stream,
@@ -70,7 +72,7 @@ func responseTencent2OpenAI(response *ChatResponse) *openai.TextResponse {
 
 func streamResponseTencent2OpenAI(TencentResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
 	response := openai.ChatCompletionsStreamResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "tencent-hunyuan",
@@ -129,7 +131,7 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 			}
 			response := streamResponseTencent2OpenAI(&TencentResponse)
 			if len(response.Choices) != 0 {
-				responseText += response.Choices[0].Delta.Content
+				responseText += conv.AsString(response.Choices[0].Delta.Content)
 			}
 			jsonResponse, err := json.Marshal(response)
 			if err != nil {

relay/adaptor/togetherai/constants.go

@@ -0,0 +1,10 @@
+package togetherai
+
+// https://docs.together.ai/docs/inference-models
+
+var ModelList = []string{
+	"meta-llama/Llama-3-70b-chat-hf",
+	"deepseek-ai/deepseek-coder-33b-instruct",
+	"mistralai/Mixtral-8x22B-Instruct-v0.1",
+	"Qwen/Qwen1.5-72B-Chat",
+}

relay/adaptor/xunfei/adaptor.go

@@ -3,10 +3,10 @@ package xunfei
 import (
 	"errors"
 	"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/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
 	"strings"
@@ -14,18 +14,27 @@ import (
 
 type Adaptor struct {
 	request *model.GeneralOpenAIRequest
+	meta    *meta.Meta
 }
 
-func (a *Adaptor) Init(meta *util.RelayMeta) {
-
+func (a *Adaptor) Init(meta *meta.Meta) {
+	a.meta = meta
 }
 
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	return "", nil
 }
 
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
+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
 }
@@ -38,14 +47,21 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return nil, nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
+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) {
 	// xunfei's request is not http request, so we don't need to do anything here
 	dummyResp := &http.Response{}
 	dummyResp.StatusCode = http.StatusOK
 	return dummyResp, nil
 }
 
-func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	splits := strings.Split(meta.APIKey, "|")
 	if len(splits) != 3 {
 		return nil, openai.ErrorWrapper(errors.New("invalid auth"), "invalid_auth", http.StatusBadRequest)
@@ -54,9 +70,9 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.Rel
 		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

@@ -11,8 +11,10 @@ import (
 	"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/channel/openai"
+	"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"
@@ -26,7 +28,11 @@ import (
 
 func requestOpenAI2Xunfei(request model.GeneralOpenAIRequest, xunfeiAppId string, domain string) *ChatRequest {
 	messages := make([]Message, 0, len(request.Messages))
+	var lastToolCalls []model.Tool
 	for _, message := range request.Messages {
+		if message.ToolCalls != nil {
+			lastToolCalls = message.ToolCalls
+		}
 		messages = append(messages, Message{
 			Role:    message.Role,
 			Content: message.StringContent(),
@@ -39,9 +45,33 @@ func requestOpenAI2Xunfei(request model.GeneralOpenAIRequest, xunfeiAppId string
 	xunfeiRequest.Parameter.Chat.TopK = request.N
 	xunfeiRequest.Parameter.Chat.MaxTokens = request.MaxTokens
 	xunfeiRequest.Payload.Message.Text = messages
+	if len(lastToolCalls) != 0 {
+		for _, toolCall := range lastToolCalls {
+			xunfeiRequest.Payload.Functions.Text = append(xunfeiRequest.Payload.Functions.Text, toolCall.Function)
+		}
+	}
+
 	return &xunfeiRequest
 }
 
+func getToolCalls(response *ChatResponse) []model.Tool {
+	var toolCalls []model.Tool
+	if len(response.Payload.Choices.Text) == 0 {
+		return toolCalls
+	}
+	item := response.Payload.Choices.Text[0]
+	if item.FunctionCall == nil {
+		return toolCalls
+	}
+	toolCall := model.Tool{
+		Id:       fmt.Sprintf("call_%s", random.GetUUID()),
+		Type:     "function",
+		Function: *item.FunctionCall,
+	}
+	toolCalls = append(toolCalls, toolCall)
+	return toolCalls
+}
+
 func responseXunfei2OpenAI(response *ChatResponse) *openai.TextResponse {
 	if len(response.Payload.Choices.Text) == 0 {
 		response.Payload.Choices.Text = []ChatResponseTextItem{
@@ -53,13 +83,14 @@ func responseXunfei2OpenAI(response *ChatResponse) *openai.TextResponse {
 	choice := openai.TextResponseChoice{
 		Index: 0,
 		Message: model.Message{
-			Role:    "assistant",
-			Content: response.Payload.Choices.Text[0].Content,
+			Role:      "assistant",
+			Content:   response.Payload.Choices.Text[0].Content,
+			ToolCalls: getToolCalls(response),
 		},
 		FinishReason: constant.StopFinishReason,
 	}
 	fullTextResponse := openai.TextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
 		Choices: []openai.TextResponseChoice{choice},
@@ -78,11 +109,12 @@ func streamResponseXunfei2OpenAI(xunfeiResponse *ChatResponse) *openai.ChatCompl
 	}
 	var choice openai.ChatCompletionsStreamResponseChoice
 	choice.Delta.Content = xunfeiResponse.Payload.Choices.Text[0].Content
+	choice.Delta.ToolCalls = getToolCalls(xunfeiResponse)
 	if xunfeiResponse.Payload.Choices.Status == 2 {
 		choice.FinishReason = &constant.StopFinishReason
 	}
 	response := openai.ChatCompletionsStreamResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
+		Id:      fmt.Sprintf("chatcmpl-%s", random.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "SparkDesk",
@@ -117,11 +149,11 @@ 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) {
-	domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model)
+func StreamHandler(c *gin.Context, meta *meta.Meta, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
+	domain, authUrl := getXunfeiAuthUrl(meta.Config.APIVersion, apiKey, apiSecret)
 	dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId)
 	if err != nil {
-		return openai.ErrorWrapper(err, "make xunfei request err", http.StatusInternalServerError), nil
+		return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil
 	}
 	common.SetEventStreamHeaders(c)
 	var usage model.Usage
@@ -147,11 +179,11 @@ 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) {
-	domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model)
+func Handler(c *gin.Context, meta *meta.Meta, textRequest model.GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*model.ErrorWithStatusCode, *model.Usage) {
+	domain, authUrl := getXunfeiAuthUrl(meta.Config.APIVersion, apiKey, apiSecret)
 	dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId)
 	if err != nil {
-		return openai.ErrorWrapper(err, "make xunfei request err", http.StatusInternalServerError), nil
+		return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil
 	}
 	var usage model.Usage
 	var content string
@@ -171,11 +203,7 @@ func Handler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId strin
 		}
 	}
 	if len(xunfeiResponse.Payload.Choices.Text) == 0 {
-		xunfeiResponse.Payload.Choices.Text = []ChatResponseTextItem{
-			{
-				Content: "",
-			},
-		}
+		return openai.ErrorWrapper(err, "xunfei_empty_response_detected", http.StatusInternalServerError), nil
 	}
 	xunfeiResponse.Payload.Choices.Text[0].Content = content
 
@@ -202,15 +230,21 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl,
 	if err != nil {
 		return nil, nil, err
 	}
+	_, msg, err := conn.ReadMessage()
+	if err != nil {
+		return nil, nil, err
+	}
 
 	dataChan := make(chan ChatResponse)
 	stopChan := make(chan bool)
 	go func() {
 		for {
-			_, msg, err := conn.ReadMessage()
-			if err != nil {
-				logger.SysError("error reading stream response: " + err.Error())
-				break
+			if msg == nil {
+				_, msg, err = conn.ReadMessage()
+				if err != nil {
+					logger.SysError("error reading stream response: " + err.Error())
+					break
+				}
 			}
 			var response ChatResponse
 			err = json.Unmarshal(msg, &response)
@@ -218,6 +252,7 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl,
 				logger.SysError("error unmarshalling stream response: " + err.Error())
 				break
 			}
+			msg = nil
 			dataChan <- response
 			if response.Payload.Choices.Status == 2 {
 				err := conn.Close()
@@ -233,25 +268,12 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl,
 	return dataChan, stopChan, nil
 }
 
-func getAPIVersion(c *gin.Context, modelName string) string {
-	query := c.Request.URL.Query()
-	apiVersion := query.Get("api-version")
-	if apiVersion != "" {
-		return apiVersion
-	}
+func parseAPIVersionByModelName(modelName string) string {
 	parts := strings.Split(modelName, "-")
 	if len(parts) == 2 {
-		apiVersion = parts[1]
-		return apiVersion
-
+		return parts[1]
 	}
-	apiVersion = c.GetString(common.ConfigKeyAPIVersion)
-	if apiVersion != "" {
-		return apiVersion
-	}
-	apiVersion = "v1.1"
-	logger.SysLog("api_version not found, using default: " + apiVersion)
-	return apiVersion
+	return ""
 }
 
 // https://www.xfyun.cn/doc/spark/Web.html#_1-%E6%8E%A5%E5%8F%A3%E8%AF%B4%E6%98%8E
@@ -269,8 +291,7 @@ func apiVersion2domain(apiVersion string) string {
 	return "general" + apiVersion
 }
 
-func getXunfeiAuthUrl(c *gin.Context, apiKey string, apiSecret string, modelName string) (string, string) {
-	apiVersion := getAPIVersion(c, modelName)
+func getXunfeiAuthUrl(apiVersion string, apiKey string, apiSecret string) (string, string) {
 	domain := apiVersion2domain(apiVersion)
 	authUrl := buildXunfeiAuthUrl(fmt.Sprintf("wss://spark-api.xf-yun.com/%s/chat", apiVersion), apiKey, apiSecret)
 	return domain, authUrl

relay/adaptor/xunfei/model.go

@@ -26,13 +26,18 @@ type ChatRequest struct {
 		Message struct {
 			Text []Message `json:"text"`
 		} `json:"message"`
+		Functions struct {
+			Text []model.Function `json:"text,omitempty"`
+		} `json:"functions,omitempty"`
 	} `json:"payload"`
 }
 
 type ChatResponseTextItem struct {
-	Content string `json:"content"`
-	Role    string `json:"role"`
-	Index   int    `json:"index"`
+	Content      string          `json:"content"`
+	Role         string          `json:"role"`
+	Index        int             `json:"index"`
+	ContentType  string          `json:"content_type"`
+	FunctionCall *model.Function `json:"function_call"`
 }
 
 type ChatResponse struct {

relay/adaptor/zhipu/adaptor.go

@@ -0,0 +1,149 @@
+package zhipu
+
+import (
+	"errors"
+	"fmt"
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/relay/relaymode"
+	"io"
+	"math"
+	"net/http"
+	"strings"
+)
+
+type Adaptor struct {
+	APIVersion string
+}
+
+func (a *Adaptor) Init(meta *meta.Meta) {
+
+}
+
+func (a *Adaptor) SetVersionByModeName(modelName string) {
+	if strings.HasPrefix(modelName, "glm-") {
+		a.APIVersion = "v4"
+	} else {
+		a.APIVersion = "v3"
+	}
+}
+
+func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
+	switch meta.Mode {
+	case relaymode.ImagesGenerations:
+		return fmt.Sprintf("%s/api/paas/v4/images/generations", meta.BaseURL), nil
+	case relaymode.Embeddings:
+		return fmt.Sprintf("%s/api/paas/v4/embeddings", meta.BaseURL), nil
+	}
+	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"
+	}
+	return fmt.Sprintf("%s/api/paas/v3/model-api/%s/%s", meta.BaseURL, meta.ActualModelName, method), nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
+	adaptor.SetupCommonRequestHeader(c, req, meta)
+	token := GetToken(meta.APIKey)
+	req.Header.Set("Authorization", token)
+	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")
+	}
+	switch relayMode {
+	case relaymode.Embeddings:
+		baiduEmbeddingRequest, err := ConvertEmbeddingRequest(*request)
+		return baiduEmbeddingRequest, err
+	default:
+		// TopP (0.0, 1.0)
+		request.TopP = math.Min(0.99, request.TopP)
+		request.TopP = math.Max(0.01, request.TopP)
+
+		// Temperature (0.0, 1.0)
+		request.Temperature = math.Min(0.99, request.Temperature)
+		request.Temperature = math.Max(0.01, request.Temperature)
+		a.SetVersionByModeName(request.Model)
+		if a.APIVersion == "v4" {
+			return request, nil
+		}
+		return ConvertRequest(*request), nil
+	}
+}
+
+func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	newRequest := ImageRequest{
+		Model:  request.Model,
+		Prompt: request.Prompt,
+		UserId: request.User,
+	}
+	return newRequest, 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) DoResponseV4(c *gin.Context, resp *http.Response, meta *meta.Meta) (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 *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	switch meta.Mode {
+	case relaymode.Embeddings:
+		err, usage = EmbeddingsHandler(c, resp)
+		return
+	case relaymode.ImagesGenerations:
+		err, usage = openai.ImageHandler(c, resp)
+		return
+	}
+	if a.APIVersion == "v4" {
+		return a.DoResponseV4(c, resp, meta)
+	}
+	if meta.IsStream {
+		err, usage = StreamHandler(c, resp)
+	} else {
+		if meta.Mode == relaymode.Embeddings {
+			err, usage = EmbeddingsHandler(c, resp)
+		} else {
+			err, usage = Handler(c, resp)
+		}
+	}
+	return
+}
+
+func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) (*EmbeddingRequest, error) {
+	inputs := request.ParseInput()
+	if len(inputs) != 1 {
+		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 {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return "zhipu"
+}

relay/adaptor/zhipu/constants.go

@@ -2,5 +2,6 @@ package zhipu
 
 var ModelList = []string{
 	"chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite",
-	"glm-4", "glm-4v", "glm-3-turbo",
+	"glm-4", "glm-4v", "glm-3-turbo", "embedding-2",
+	"cogview-3",
 }

relay/adaptor/zhipu/main.go

@@ -8,7 +8,7 @@ import (
 	"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/channel/openai"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/model"
 	"io"
@@ -254,3 +254,50 @@ func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *
 	_, err = c.Writer.Write(jsonResponse)
 	return nil, &fullTextResponse.Usage
 }
+
+func EmbeddingsHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	var zhipuResponse EmbeddingResponse
+	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
+	}
+	err = json.Unmarshal(responseBody, &zhipuResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	fullTextResponse := embeddingResponseZhipu2OpenAI(&zhipuResponse)
+	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, &fullTextResponse.Usage
+}
+
+func embeddingResponseZhipu2OpenAI(response *EmbeddingResponse) *openai.EmbeddingResponse {
+	openAIEmbeddingResponse := openai.EmbeddingResponse{
+		Object: "list",
+		Data:   make([]openai.EmbeddingResponseItem, 0, len(response.Embeddings)),
+		Model:  response.Model,
+		Usage: model.Usage{
+			PromptTokens:     response.PromptTokens,
+			CompletionTokens: response.CompletionTokens,
+			TotalTokens:      response.Usage.TotalTokens,
+		},
+	}
+
+	for _, item := range response.Embeddings {
+		openAIEmbeddingResponse.Data = append(openAIEmbeddingResponse.Data, openai.EmbeddingResponseItem{
+			Object:    `embedding`,
+			Index:     item.Index,
+			Embedding: item.Embedding,
+		})
+	}
+	return &openAIEmbeddingResponse
+}

relay/adaptor/zhipu/model.go

@@ -44,3 +44,27 @@ type tokenData struct {
 	Token      string
 	ExpiryTime time.Time
 }
+
+type EmbeddingRequest struct {
+	Model string `json:"model"`
+	Input string `json:"input"`
+}
+
+type EmbeddingResponse struct {
+	Model       string          `json:"model"`
+	Object      string          `json:"object"`
+	Embeddings  []EmbeddingData `json:"data"`
+	model.Usage `json:"usage"`
+}
+
+type EmbeddingData struct {
+	Index     int       `json:"index"`
+	Object    string    `json:"object"`
+	Embedding []float64 `json:"embedding"`
+}
+
+type ImageRequest struct {
+	Model  string `json:"model"`
+	Prompt string `json:"prompt"`
+	UserId string `json:"user_id,omitempty"`
+}

relay/apitype/define.go

@@ -0,0 +1,22 @@
+package apitype
+
+const (
+	OpenAI = iota
+	Anthropic
+	PaLM
+	Baidu
+	Zhipu
+	Ali
+	Xunfei
+	AIProxyLibrary
+	Tencent
+	Gemini
+	Ollama
+	AwsClaude
+	Coze
+	Cohere
+	Cloudflare
+	DeepL
+
+	Dummy // this one is only for count, do not add any channel after this
+)

relay/billing/billing.go

@@ -0,0 +1,42 @@
+package billing
+
+import (
+	"context"
+	"fmt"
+	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/model"
+)
+
+func ReturnPreConsumedQuota(ctx context.Context, preConsumedQuota int64, tokenId int) {
+	if preConsumedQuota != 0 {
+		go func(ctx context.Context) {
+			// return pre-consumed quota
+			err := model.PostConsumeTokenQuota(tokenId, -preConsumedQuota)
+			if err != nil {
+				logger.Error(ctx, "error return pre-consumed quota: "+err.Error())
+			}
+		}(ctx)
+	}
+}
+
+func PostConsumeQuota(ctx context.Context, tokenId int, quotaDelta int64, totalQuota int64, userId int, channelId int, modelRatio float64, groupRatio float64, modelName string, tokenName string) {
+	// quotaDelta is remaining quota to be consumed
+	err := model.PostConsumeTokenQuota(tokenId, quotaDelta)
+	if err != nil {
+		logger.SysError("error consuming token remain quota: " + err.Error())
+	}
+	err = model.CacheUpdateUserQuota(ctx, userId)
+	if err != nil {
+		logger.SysError("error update user quota cache: " + err.Error())
+	}
+	// totalQuota is total quota consumed
+	if totalQuota != 0 {
+		logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
+		model.RecordConsumeLog(ctx, userId, channelId, int(totalQuota), 0, modelName, tokenName, totalQuota, logContent)
+		model.UpdateUserUsedQuotaAndRequestCount(userId, totalQuota)
+		model.UpdateChannelUsedQuota(channelId, totalQuota)
+	}
+	if totalQuota <= 0 {
+		logger.Error(ctx, fmt.Sprintf("totalQuota consumed is %d, something is wrong", totalQuota))
+	}
+}

relay/billing/ratio/group.go

@@ -0,0 +1,34 @@
+package ratio
+
+import (
+	"encoding/json"
+	"github.com/songquanpeng/one-api/common/logger"
+)
+
+var GroupRatio = map[string]float64{
+	"default": 1,
+	"vip":     1,
+	"svip":    1,
+}
+
+func GroupRatio2JSONString() string {
+	jsonBytes, err := json.Marshal(GroupRatio)
+	if err != nil {
+		logger.SysError("error marshalling model ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+func UpdateGroupRatioByJSONString(jsonStr string) error {
+	GroupRatio = make(map[string]float64)
+	return json.Unmarshal([]byte(jsonStr), &GroupRatio)
+}
+
+func GetGroupRatio(name string) float64 {
+	ratio, ok := GroupRatio[name]
+	if !ok {
+		logger.SysError("group ratio not found: " + name)
+		return 1
+	}
+	return ratio
+}

relay/billing/ratio/image.go

@@ -0,0 +1,51 @@
+package ratio
+
+var ImageSizeRatios = 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,
+	},
+	"ali-stable-diffusion-xl": {
+		"512x1024":  1,
+		"1024x768":  1,
+		"1024x1024": 1,
+		"576x1024":  1,
+		"1024x576":  1,
+	},
+	"ali-stable-diffusion-v1.5": {
+		"512x1024":  1,
+		"1024x768":  1,
+		"1024x1024": 1,
+		"576x1024":  1,
+		"1024x576":  1,
+	},
+	"wanx-v1": {
+		"1024x1024": 1,
+		"720x1280":  1,
+		"1280x720":  1,
+	},
+}
+
+var ImageGenerationAmounts = map[string][2]int{
+	"dall-e-2":                  {1, 10},
+	"dall-e-3":                  {1, 1}, // OpenAI allows n=1 currently.
+	"ali-stable-diffusion-xl":   {1, 4}, // Ali
+	"ali-stable-diffusion-v1.5": {1, 4}, // Ali
+	"wanx-v1":                   {1, 4}, // Ali
+	"cogview-3":                 {1, 1},
+}
+
+var ImagePromptLengthLimitations = map[string]int{
+	"dall-e-2":                  1000,
+	"dall-e-3":                  4000,
+	"ali-stable-diffusion-xl":   4000,
+	"ali-stable-diffusion-v1.5": 4000,
+	"wanx-v1":                   4000,
+	"cogview-3":                 833,
+}

relay/billing/ratio/model.go

@@ -0,0 +1,323 @@
+package ratio
+
+import (
+	"encoding/json"
+	"strings"
+
+	"github.com/songquanpeng/one-api/common/logger"
+)
+
+const (
+	USD2RMB = 7
+	USD     = 500 // $0.002 = 1 -> $1 = 500
+	RMB     = USD / USD2RMB
+)
+
+// ModelRatio
+// https://platform.openai.com/docs/models/model-endpoint-compatibility
+// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
+// https://openai.com/pricing
+// 1 === $0.002 / 1K tokens
+// 1 === ￥0.014 / 1k tokens
+var ModelRatio = map[string]float64{
+
+	// https://openai.com/pricing
+	"gpt-4":                   15,
+	"gpt-4-0314":              15,
+	"gpt-4-0613":              15,
+  "gpt-4-gizmo":             15,
+	"gpt-4-32k":               30,
+	"gpt-4-32k-0314":          30,
+	"gpt-4-32k-0613":          30,
+	"gpt-4-1106-preview":      5,    // $0.01 / 1K tokens
+	"gpt-4-0125-preview":      5,    // $0.01 / 1K tokens
+	"gpt-4-turbo-preview":     5,    // $0.01 / 1K tokens
+	"gpt-4-turbo":             5,    // $0.01 / 1K tokens
+	"gpt-4-turbo-2024-04-09":  5,    // $0.01 / 1K tokens
+	"gpt-4-vision-preview":    5,    // $0.01 / 1K tokens
+	"gpt-3.5-turbo":           0.25, // $0.0005 / 1K tokens
+	"gpt-3.5-turbo-0301":      0.75,
+	"gpt-3.5-turbo-0613":      0.75,
+	"gpt-3.5-turbo-16k":       1.5, // $0.003 / 1K tokens
+	"gpt-3.5-turbo-16k-0613":  1.5,
+	"gpt-3.5-turbo-instruct":  0.75, // $0.0015 / 1K tokens
+	"gpt-3.5-turbo-1106":      0.5,  // $0.001 / 1K tokens
+	"gpt-3.5-turbo-0125":      0.25, // $0.0005 / 1K tokens
+	"davinci-002":             1,    // $0.002 / 1K tokens
+	"babbage-002":             0.2,  // $0.0004 / 1K tokens
+	"text-ada-001":            0.2,
+	"text-babbage-001":        0.25,
+	"text-curie-001":          1,
+	"text-davinci-002":        10,
+	"text-davinci-003":        10,
+	"text-davinci-edit-001":   10,
+	"code-davinci-edit-001":   10,
+	"whisper-1":               15,  // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
+	"tts-1":                   7.5, // $0.015 / 1K characters
+	"tts-1-1106":              7.5,
+	"tts-1-hd":                15, // $0.030 / 1K characters
+	"tts-1-hd-1106":           15,
+	"davinci":                 10,
+	"curie":                   10,
+	"babbage":                 10,
+	"ada":                     10,
+	"text-embedding-ada-002":  0.05,
+	"text-embedding-3-small":  0.01,
+	"text-embedding-3-large":  0.065,
+	"text-search-ada-doc-001": 10,
+	"text-moderation-stable":  0.1,
+	"text-moderation-latest":  0.1,
+	"dall-e-2":                0.02 * USD, // $0.016 - $0.020 / image
+	"dall-e-3":                0.04 * USD, // $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-20240307":  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-4.0-8K":       0.120 * RMB,
+	"ERNIE-3.5-8K":       0.012 * RMB,
+	"ERNIE-3.5-8K-0205":  0.024 * RMB,
+	"ERNIE-3.5-8K-1222":  0.012 * RMB,
+	"ERNIE-Bot-8K":       0.024 * RMB,
+	"ERNIE-3.5-4K-0205":  0.012 * RMB,
+	"ERNIE-Speed-8K":     0.004 * RMB,
+	"ERNIE-Speed-128K":   0.004 * RMB,
+	"ERNIE-Lite-8K-0922": 0.008 * RMB,
+	"ERNIE-Lite-8K-0308": 0.003 * RMB,
+	"ERNIE-Tiny-8K":      0.001 * RMB,
+	"BLOOMZ-7B":          0.004 * RMB,
+	"Embedding-V1":       0.002 * RMB,
+	"bge-large-zh":       0.002 * RMB,
+	"bge-large-en":       0.002 * RMB,
+	"tao-8k":             0.002 * RMB,
+	// https://ai.google.dev/pricing
+	"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
+	"gemini-1.0-pro-vision-001": 1,
+	"gemini-1.0-pro-001":        1,
+	"gemini-1.5-pro":            1,
+	// https://open.bigmodel.cn/pricing
+	"glm-4":         0.1 * RMB,
+	"glm-4v":        0.1 * RMB,
+	"glm-3-turbo":   0.005 * RMB,
+	"embedding-2":   0.0005 * 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
+	"chatglm_lite":  0.1429, // ￥0.002 / 1k tokens
+	"cogview-3":     0.25 * RMB,
+	// https://help.aliyun.com/zh/dashscope/developer-reference/tongyi-thousand-questions-metering-and-billing
+	"qwen-turbo":                0.5715, // ￥0.008 / 1k tokens
+	"qwen-plus":                 1.4286, // ￥0.02 / 1k tokens
+	"qwen-max":                  1.4286, // ￥0.02 / 1k tokens
+	"qwen-max-longcontext":      1.4286, // ￥0.02 / 1k tokens
+	"text-embedding-v1":         0.05,   // ￥0.0007 / 1k tokens
+	"ali-stable-diffusion-xl":   8,
+	"ali-stable-diffusion-v1.5": 8,
+	"wanx-v1":                   8,
+	"SparkDesk":                 1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v1.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v2.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.5":            1.2858, // ￥0.018 / 1k tokens
+	"360GPT_S2_V9":              0.8572, // ¥0.012 / 1k tokens
+	"embedding-bert-512-v1":     0.0715, // ¥0.001 / 1k tokens
+	"embedding_s1_v1":           0.0715, // ¥0.001 / 1k tokens
+	"semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
+	"hunyuan":                   7.143,  // ¥0.1 / 1k tokens  // https://cloud.tencent.com/document/product/1729/97731#e0e6be58-60c8-469f-bdeb-6c264ce3b4d0
+	"ChatStd":                   0.01 * RMB,
+	"ChatPro":                   0.1 * RMB,
+	// https://platform.moonshot.cn/pricing
+	"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
+	"abab6.5-chat":  0.03 * RMB,
+	"abab6.5s-chat": 0.01 * RMB,
+	"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,
+	// https://wow.groq.com/#:~:text=inquiries%C2%A0here.-,Model,-Current%20Speed
+	"llama3-70b-8192":    0.59 / 1000 * USD,
+	"mixtral-8x7b-32768": 0.27 / 1000 * USD,
+	"llama3-8b-8192":     0.05 / 1000 * USD,
+	"gemma-7b-it":        0.1 / 1000 * USD,
+	"llama2-70b-4096":    0.64 / 1000 * USD,
+	"llama2-7b-2048":     0.1 / 1000 * USD,
+	// https://platform.lingyiwanwu.com/docs#-计费单元
+	"yi-34b-chat-0205": 2.5 / 1000 * RMB,
+	"yi-34b-chat-200k": 12.0 / 1000 * RMB,
+	"yi-vl-plus":       6.0 / 1000 * RMB,
+	// stepfun todo
+	"step-1v-32k": 0.024 * RMB,
+	"step-1-32k":  0.024 * RMB,
+	"step-1-200k": 0.15 * RMB,
+	// https://cohere.com/pricing
+	"command":               0.5,
+	"command-nightly":       0.5,
+	"command-light":         0.5,
+	"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{}
+
+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 AddNewMissingRatio(oldRatio string) string {
+	newRatio := make(map[string]float64)
+	err := json.Unmarshal([]byte(oldRatio), &newRatio)
+	if err != nil {
+		logger.SysError("error unmarshalling old ratio: " + err.Error())
+		return oldRatio
+	}
+	for k, v := range DefaultModelRatio {
+		if _, ok := newRatio[k]; !ok {
+			newRatio[k] = v
+		}
+	}
+	jsonBytes, err := json.Marshal(newRatio)
+	if err != nil {
+		logger.SysError("error marshalling new ratio: " + err.Error())
+		return oldRatio
+	}
+	return string(jsonBytes)
+}
+
+func ModelRatio2JSONString() string {
+	jsonBytes, err := json.Marshal(ModelRatio)
+	if err != nil {
+		logger.SysError("error marshalling model ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+func UpdateModelRatioByJSONString(jsonStr string) error {
+	ModelRatio = make(map[string]float64)
+	return json.Unmarshal([]byte(jsonStr), &ModelRatio)
+}
+
+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 strings.Index(name, "gpt-4-gizmo") != -1 {
+		return ModelRatio["gpt-4-gizmo"]
+	}
+	if !ok {
+		ratio, ok = DefaultModelRatio[name]
+	}
+	if !ok {
+		logger.SysError("model ratio not found: " + name)
+		return 30
+	}
+	return ratio
+}
+
+func CompletionRatio2JSONString() string {
+	jsonBytes, err := json.Marshal(CompletionRatio)
+	if err != nil {
+		logger.SysError("error marshalling completion ratio: " + err.Error())
+	}
+	return string(jsonBytes)
+}
+
+func UpdateCompletionRatioByJSONString(jsonStr string) error {
+	CompletionRatio = make(map[string]float64)
+	return json.Unmarshal([]byte(jsonStr), &CompletionRatio)
+}
+
+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 name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
+			// https://openai.com/blog/new-embedding-models-and-api-updates
+			// Updated GPT-3.5 Turbo model and lower pricing
+			return 3
+		}
+		if strings.HasSuffix(name, "1106") {
+			return 2
+		}
+		return 4.0 / 3.0
+	}
+	if strings.HasPrefix(name, "gpt-4") {
+		if strings.HasPrefix(name, "gpt-4-turbo") || strings.HasSuffix(name, "preview") {
+			return 3
+		}
+		return 2
+	}
+	if strings.HasPrefix(name, "claude-3") {
+		return 5
+	}
+	if strings.HasPrefix(name, "claude-") {
+		return 3
+	}
+	if strings.HasPrefix(name, "mistral-") {
+		return 3
+	}
+	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
+	case "llama3-8b-8192":
+		return 2
+	case "llama3-70b-8192":
+		return 0.79 / 0.59
+	case "command", "command-light", "command-nightly", "command-light-nightly":
+		return 2
+	case "command-r":
+		return 3
+	case "command-r-plus":
+		return 5
+	}
+	return 1
+}

relay/channel/ali/adaptor.go

@@ -1,86 +0,0 @@
-package ali
-
-import (
-	"errors"
-	"fmt"
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/common"
-	"github.com/songquanpeng/one-api/relay/channel"
-	"github.com/songquanpeng/one-api/relay/constant"
-	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
-	"io"
-	"net/http"
-)
-
-// https://help.aliyun.com/zh/dashscope/developer-reference/api-details
-
-type Adaptor struct {
-}
-
-func (a *Adaptor) Init(meta *util.RelayMeta) {
-
-}
-
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	fullRequestURL := fmt.Sprintf("%s/api/v1/services/aigc/text-generation/generation", meta.BaseURL)
-	if meta.Mode == constant.RelayModeEmbeddings {
-		fullRequestURL = fmt.Sprintf("%s/api/v1/services/embeddings/text-embedding/text-embedding", meta.BaseURL)
-	}
-	return fullRequestURL, nil
-}
-
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.SetupCommonRequestHeader(c, req, meta)
-	if meta.IsStream {
-		req.Header.Set("Accept", "text/event-stream")
-	}
-	req.Header.Set("Authorization", "Bearer "+meta.APIKey)
-	if meta.IsStream {
-		req.Header.Set("X-DashScope-SSE", "enable")
-	}
-	if c.GetString(common.ConfigKeyPlugin) != "" {
-		req.Header.Set("X-DashScope-Plugin", c.GetString(common.ConfigKeyPlugin))
-	}
-	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")
-	}
-	switch relayMode {
-	case constant.RelayModeEmbeddings:
-		baiduEmbeddingRequest := ConvertEmbeddingRequest(*request)
-		return baiduEmbeddingRequest, nil
-	default:
-		baiduRequest := ConvertRequest(*request)
-		return baiduRequest, nil
-	}
-}
-
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
-}
-
-func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
-	if meta.IsStream {
-		err, usage = StreamHandler(c, resp)
-	} else {
-		switch meta.Mode {
-		case constant.RelayModeEmbeddings:
-			err, usage = EmbeddingHandler(c, resp)
-		default:
-			err, usage = Handler(c, resp)
-		}
-	}
-	return
-}
-
-func (a *Adaptor) GetModelList() []string {
-	return ModelList
-}
-
-func (a *Adaptor) GetChannelName() string {
-	return "ali"
-}

relay/channel/ali/model.go

@@ -1,73 +0,0 @@
-package ali
-
-type Message struct {
-	Content string `json:"content"`
-	Role    string `json:"role"`
-}
-
-type Input struct {
-	//Prompt   string       `json:"prompt"`
-	Messages []Message `json:"messages"`
-}
-
-type Parameters struct {
-	TopP              float64 `json:"top_p,omitempty"`
-	TopK              int     `json:"top_k,omitempty"`
-	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 {
-	Model      string     `json:"model"`
-	Input      Input      `json:"input"`
-	Parameters Parameters `json:"parameters,omitempty"`
-}
-
-type EmbeddingRequest struct {
-	Model string `json:"model"`
-	Input struct {
-		Texts []string `json:"texts"`
-	} `json:"input"`
-	Parameters *struct {
-		TextType string `json:"text_type,omitempty"`
-	} `json:"parameters,omitempty"`
-}
-
-type Embedding struct {
-	Embedding []float64 `json:"embedding"`
-	TextIndex int       `json:"text_index"`
-}
-
-type EmbeddingResponse struct {
-	Output struct {
-		Embeddings []Embedding `json:"embeddings"`
-	} `json:"output"`
-	Usage Usage `json:"usage"`
-	Error
-}
-
-type Error struct {
-	Code      string `json:"code"`
-	Message   string `json:"message"`
-	RequestId string `json:"request_id"`
-}
-
-type Usage struct {
-	InputTokens  int `json:"input_tokens"`
-	OutputTokens int `json:"output_tokens"`
-	TotalTokens  int `json:"total_tokens"`
-}
-
-type Output struct {
-	Text         string `json:"text"`
-	FinishReason string `json:"finish_reason"`
-}
-
-type ChatResponse struct {
-	Output Output `json:"output"`
-	Usage  Usage  `json:"usage"`
-	Error
-}

relay/channel/baidu/constants.go

@@ -1,13 +0,0 @@
-package baidu
-
-var ModelList = []string{
-	"ERNIE-Bot-4",
-	"ERNIE-Bot-8K",
-	"ERNIE-Bot",
-	"ERNIE-Speed",
-	"ERNIE-Bot-turbo",
-	"Embedding-V1",
-	"bge-large-zh",
-	"bge-large-en",
-	"tao-8k",
-}

relay/channel/interface.go

@@ -1,20 +0,0 @@
-package channel
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
-	"io"
-	"net/http"
-)
-
-type Adaptor interface {
-	Init(meta *util.RelayMeta)
-	GetRequestURL(meta *util.RelayMeta) (string, error)
-	SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error
-	ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error)
-	DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error)
-	DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode)
-	GetModelList() []string
-	GetChannelName() string
-}

relay/channel/minimax/constants.go

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

relay/channel/minimax/main.go

@@ -1,14 +0,0 @@
-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/ollama/adaptor.go

@@ -1,65 +0,0 @@
-package ollama
-
-import (
-	"errors"
-	"fmt"
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/channel"
-	"github.com/songquanpeng/one-api/relay/constant"
-	"github.com/songquanpeng/one-api/relay/model"
-	"github.com/songquanpeng/one-api/relay/util"
-	"io"
-	"net/http"
-)
-
-type Adaptor struct {
-}
-
-func (a *Adaptor) Init(meta *util.RelayMeta) {
-
-}
-
-func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	// https://github.com/ollama/ollama/blob/main/docs/api.md
-	fullRequestURL := fmt.Sprintf("%s/api/chat", meta.BaseURL)
-	return fullRequestURL, nil
-}
-
-func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
-	channel.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")
-	}
-	switch relayMode {
-	case constant.RelayModeEmbeddings:
-		return nil, errors.New("not supported")
-	default:
-		return ConvertRequest(*request), nil
-	}
-}
-
-func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io.Reader) (*http.Response, error) {
-	return channel.DoRequestHelper(a, c, meta, requestBody)
-}
-
-func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
-	if meta.IsStream {
-		err, usage = StreamHandler(c, resp)
-	} else {
-		err, usage = Handler(c, resp)
-	}
-	return
-}
-
-func (a *Adaptor) GetModelList() []string {
-	return ModelList
-}
-
-func (a *Adaptor) GetChannelName() string {
-	return "ollama"
-}