feat: load channel models from server

百度千帆平台上的第三方大模型调用

README.md

@@ -67,6 +67,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [OpenAI ChatGPT 系列模型](https://platform.openai.com/docs/guides/gpt/chat-completions-api)（支持 [Azure OpenAI API](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference)）
    + [x] [Anthropic Claude 系列模型](https://anthropic.com)
    + [x] [Google PaLM2/Gemini 系列模型](https://developers.generativeai.google)
+   + [x] [Mistral 系列模型](https://mistral.ai/)
    + [x] [百度文心一言系列模型](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
    + [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
    + [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)

common/constants.go

@@ -38,34 +38,37 @@ const (
 )
 
 const (
-	ChannelTypeUnknown        = 0
-	ChannelTypeOpenAI         = 1
-	ChannelTypeAPI2D          = 2
-	ChannelTypeAzure          = 3
-	ChannelTypeCloseAI        = 4
-	ChannelTypeOpenAISB       = 5
-	ChannelTypeOpenAIMax      = 6
-	ChannelTypeOhMyGPT        = 7
-	ChannelTypeCustom         = 8
-	ChannelTypeAILS           = 9
-	ChannelTypeAIProxy        = 10
-	ChannelTypePaLM           = 11
-	ChannelTypeAPI2GPT        = 12
-	ChannelTypeAIGC2D         = 13
-	ChannelTypeAnthropic      = 14
-	ChannelTypeBaidu          = 15
-	ChannelTypeZhipu          = 16
-	ChannelTypeAli            = 17
-	ChannelTypeXunfei         = 18
-	ChannelType360            = 19
-	ChannelTypeOpenRouter     = 20
-	ChannelTypeAIProxyLibrary = 21
-	ChannelTypeFastGPT        = 22
-	ChannelTypeTencent        = 23
-	ChannelTypeGemini         = 24
-	ChannelTypeMoonshot       = 25
-	ChannelTypeBaichuan       = 26
-	ChannelTypeMinimax        = 27
+	ChannelTypeUnknown = iota
+	ChannelTypeOpenAI
+	ChannelTypeAPI2D
+	ChannelTypeAzure
+	ChannelTypeCloseAI
+	ChannelTypeOpenAISB
+	ChannelTypeOpenAIMax
+	ChannelTypeOhMyGPT
+	ChannelTypeCustom
+	ChannelTypeAILS
+	ChannelTypeAIProxy
+	ChannelTypePaLM
+	ChannelTypeAPI2GPT
+	ChannelTypeAIGC2D
+	ChannelTypeAnthropic
+	ChannelTypeBaidu
+	ChannelTypeZhipu
+	ChannelTypeAli
+	ChannelTypeXunfei
+	ChannelType360
+	ChannelTypeOpenRouter
+	ChannelTypeAIProxyLibrary
+	ChannelTypeFastGPT
+	ChannelTypeTencent
+	ChannelTypeGemini
+	ChannelTypeMoonshot
+	ChannelTypeBaichuan
+	ChannelTypeMinimax
+	ChannelTypeMistral
+
+	ChannelTypeDummy
 )
 
 var ChannelBaseURLs = []string{
@@ -97,6 +100,7 @@ var ChannelBaseURLs = []string{
 	"https://api.moonshot.cn",                   // 25
 	"https://api.baichuan-ai.com",               // 26
 	"https://api.minimax.chat",                  // 27
+	"https://api.mistral.ai",                    // 28
 }
 
 const (

common/model-ratio.go

@@ -17,7 +17,6 @@ const (
 // https://platform.openai.com/docs/models/model-endpoint-compatibility
 // https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
 // https://openai.com/pricing
-// TODO: when a new api is enabled, check the pricing here
 // 1 === $0.002 / 1K tokens
 // 1 === ￥0.014 / 1k tokens
 var ModelRatio = map[string]float64{
@@ -64,12 +63,15 @@ var ModelRatio = map[string]float64{
 	"text-search-ada-doc-001": 10,
 	"text-moderation-stable":  0.1,
 	"text-moderation-latest":  0.1,
-	"dall-e-2":                8,     // $0.016 - $0.020 / image
-	"dall-e-3":                20,    // $0.040 - $0.120 / image
-	"claude-instant-1":        0.815, // $1.63 / 1M tokens
-	"claude-2":                5.51,  // $11.02 / 1M tokens
-	"claude-2.0":              5.51,  // $11.02 / 1M tokens
-	"claude-2.1":              5.51,  // $11.02 / 1M tokens
+	"dall-e-2":                8,  // $0.016 - $0.020 / image
+	"dall-e-3":                20, // $0.040 - $0.120 / image
+	// https://www.anthropic.com/api#pricing
+	"claude-instant-1.2":       0.8 / 1000 * USD,
+	"claude-2.0":               8.0 / 1000 * USD,
+	"claude-2.1":               8.0 / 1000 * USD,
+	"claude-3-haiku-20240229":  0.25 / 1000 * USD,
+	"claude-3-sonnet-20240229": 3.0 / 1000 * USD,
+	"claude-3-opus-20240229":   15.0 / 1000 * USD,
 	// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7
 	"ERNIE-Bot":         0.8572,     // ￥0.012 / 1k tokens
 	"ERNIE-Bot-turbo":   0.5715,     // ￥0.008 / 1k tokens
@@ -116,15 +118,29 @@ var ModelRatio = map[string]float64{
 	"abab6-chat":    0.1 * RMB,
 	"abab5.5-chat":  0.015 * RMB,
 	"abab5.5s-chat": 0.005 * RMB,
+	// https://docs.mistral.ai/platform/pricing/
+	"open-mistral-7b":       0.25 / 1000 * USD,
+	"open-mixtral-8x7b":     0.7 / 1000 * USD,
+	"mistral-small-latest":  2.0 / 1000 * USD,
+	"mistral-medium-latest": 2.7 / 1000 * USD,
+	"mistral-large-latest":  8.0 / 1000 * USD,
+	"mistral-embed":         0.1 / 1000 * USD,
 }
 
+var CompletionRatio = map[string]float64{}
+
 var DefaultModelRatio map[string]float64
+var DefaultCompletionRatio map[string]float64
 
 func init() {
 	DefaultModelRatio = make(map[string]float64)
 	for k, v := range ModelRatio {
 		DefaultModelRatio[k] = v
 	}
+	DefaultCompletionRatio = make(map[string]float64)
+	for k, v := range CompletionRatio {
+		DefaultCompletionRatio[k] = v
+	}
 }
 
 func ModelRatio2JSONString() string {
@@ -155,8 +171,6 @@ func GetModelRatio(name string) float64 {
 	return ratio
 }
 
-var CompletionRatio = map[string]float64{}
-
 func CompletionRatio2JSONString() string {
 	jsonBytes, err := json.Marshal(CompletionRatio)
 	if err != nil {
@@ -174,6 +188,9 @@ func GetCompletionRatio(name string) float64 {
 	if ratio, ok := CompletionRatio[name]; ok {
 		return ratio
 	}
+	if ratio, ok := DefaultCompletionRatio[name]; ok {
+		return ratio
+	}
 	if strings.HasPrefix(name, "gpt-3.5") {
 		if strings.HasSuffix(name, "0125") {
 			// https://openai.com/blog/new-embedding-models-and-api-updates
@@ -200,11 +217,14 @@ func GetCompletionRatio(name string) float64 {
 		}
 		return 2
 	}
-	if strings.HasPrefix(name, "claude-instant-1") {
-		return 3.38
+	if strings.HasPrefix(name, "claude-3") {
+		return 5
 	}
-	if strings.HasPrefix(name, "claude-2") {
-		return 2.965517
+	if strings.HasPrefix(name, "claude-") {
+		return 3
+	}
+	if strings.HasPrefix(name, "mistral-") {
+		return 3
 	}
 	return 1
 }

common/random.go

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

controller/channel-test.go

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

controller/model.go

@@ -3,13 +3,17 @@ package controller
 import (
 	"fmt"
 	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
 	"github.com/songquanpeng/one-api/relay/channel/baichuan"
 	"github.com/songquanpeng/one-api/relay/channel/minimax"
+	"github.com/songquanpeng/one-api/relay/channel/mistral"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/helper"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/relay/util"
+	"net/http"
 )
 
 // https://platform.openai.com/docs/api-reference/models/list
@@ -41,6 +45,7 @@ type OpenAIModels struct {
 
 var openAIModels []OpenAIModels
 var openAIModelsMap map[string]OpenAIModels
+var channelId2Models map[int][]string
 
 func init() {
 	var permission []OpenAIModelPermission
@@ -122,10 +127,38 @@ func init() {
 			Parent:     nil,
 		})
 	}
+	for _, modelName := range mistral.ModelList {
+		openAIModels = append(openAIModels, OpenAIModels{
+			Id:         modelName,
+			Object:     "model",
+			Created:    1626777600,
+			OwnedBy:    "mistralai",
+			Permission: permission,
+			Root:       modelName,
+			Parent:     nil,
+		})
+	}
 	openAIModelsMap = make(map[string]OpenAIModels)
 	for _, model := range openAIModels {
 		openAIModelsMap[model.Id] = model
 	}
+	channelId2Models = make(map[int][]string)
+	for i := 1; i < common.ChannelTypeDummy; i++ {
+		adaptor := helper.GetAdaptor(constant.ChannelType2APIType(i))
+		meta := &util.RelayMeta{
+			ChannelType: i,
+		}
+		adaptor.Init(meta)
+		channelId2Models[i] = adaptor.GetModelList()
+	}
+}
+
+func DashboardListModels(c *gin.Context) {
+	c.JSON(http.StatusOK, gin.H{
+		"success": true,
+		"message": "",
+		"data":    channelId2Models,
+	})
 }
 
 func ListModels(c *gin.Context) {

controller/relay.go

@@ -62,7 +62,7 @@ func Relay(c *gin.Context) {
 		retryTimes = 0
 	}
 	for i := retryTimes; i > 0; i-- {
-		channel, err := dbmodel.CacheGetRandomSatisfiedChannel(group, originalModel)
+		channel, err := dbmodel.CacheGetRandomSatisfiedChannel(group, originalModel, i != retryTimes)
 		if err != nil {
 			logger.Errorf(ctx, "CacheGetRandomSatisfiedChannel failed: %w", err)
 			break

middleware/distributor.go

@@ -68,7 +68,7 @@ func Distribute() func(c *gin.Context) {
 				}
 			}
 			requestModel = modelRequest.Model
-			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model)
+			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model, false)
 			if err != nil {
 				message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", userGroup, modelRequest.Model)
 				if channel != nil {

model/cache.go

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

relay/channel/ali/main.go

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

relay/channel/ali/model.go

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

relay/channel/anthropic/adaptor.go

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

relay/channel/anthropic/constants.go

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

relay/channel/anthropic/main.go

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

relay/channel/anthropic/model.go

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

relay/channel/baidu/adaptor.go

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

relay/channel/gemini/constants.go

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

relay/channel/mistral/constants.go

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

relay/channel/openai/adaptor.go

@@ -9,6 +9,7 @@ import (
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
 	"github.com/songquanpeng/one-api/relay/channel/baichuan"
 	"github.com/songquanpeng/one-api/relay/channel/minimax"
+	"github.com/songquanpeng/one-api/relay/channel/mistral"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
@@ -94,6 +95,8 @@ func (a *Adaptor) GetModelList() []string {
 		return baichuan.ModelList
 	case common.ChannelTypeMinimax:
 		return minimax.ModelList
+	case common.ChannelTypeMistral:
+		return mistral.ModelList
 	default:
 		return ModelList
 	}
@@ -111,6 +114,8 @@ func (a *Adaptor) GetChannelName() string {
 		return "baichuan"
 	case common.ChannelTypeMinimax:
 		return "minimax"
+	case common.ChannelTypeMistral:
+		return "mistralai"
 	default:
 		return "openai"
 	}

relay/controller/helper.go

@@ -172,10 +172,8 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *util.R
 	if err != nil {
 		logger.Error(ctx, "error update user quota cache: "+err.Error())
 	}
-	if quota != 0 {
-		logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
-		model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
-		model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
-		model.UpdateChannelUsedQuota(meta.ChannelId, quota)
-	}
+	logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
+	model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
+	model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
+	model.UpdateChannelUsedQuota(meta.ChannelId, quota)
 }

router/api-router.go

@@ -14,6 +14,7 @@ func SetApiRouter(router *gin.Engine) {
 	apiRouter.Use(middleware.GlobalAPIRateLimit())
 	{
 		apiRouter.GET("/status", controller.GetStatus)
+		apiRouter.GET("/models", middleware.UserAuth(), controller.DashboardListModels)
 		apiRouter.GET("/notice", controller.GetNotice)
 		apiRouter.GET("/about", controller.GetAbout)
 		apiRouter.GET("/home_page_content", controller.GetHomePageContent)

web/berry/src/constants/ChannelConstants.js

@@ -29,6 +29,12 @@ export const CHANNEL_OPTIONS = {
     value: 24,
     color: 'orange'
   },
+  28: {
+    key: 28,
+    text: 'Mistral AI',
+    value: 28,
+    color: 'orange'
+  },
   15: {
     key: 15,
     text: '百度文心千帆',

web/default/src/components/ChannelsTable.js

@@ -1,7 +1,16 @@
 import React, { useEffect, useState } from 'react';
 import { Button, Form, Input, Label, Message, Pagination, Popup, Table } from 'semantic-ui-react';
 import { Link } from 'react-router-dom';
-import { API, setPromptShown, shouldShowPrompt, showError, showInfo, showSuccess, timestamp2string } from '../helpers';
+import {
+  API,
+  loadChannelModels,
+  setPromptShown,
+  shouldShowPrompt,
+  showError,
+  showInfo,
+  showSuccess,
+  timestamp2string
+} from '../helpers';
 
 import { CHANNEL_OPTIONS, ITEMS_PER_PAGE } from '../constants';
 import { renderGroup, renderNumber } from '../helpers/render';
@@ -95,6 +104,7 @@ const ChannelsTable = () => {
       .catch((reason) => {
         showError(reason);
       });
+    loadChannelModels().then();
   }, []);
 
   const manageChannel = async (id, action, idx, value) => {

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

@@ -4,6 +4,7 @@ export const CHANNEL_OPTIONS = [
   { key: 3, text: 'Azure OpenAI', value: 3, color: 'olive' },
   { key: 11, text: 'Google PaLM2', value: 11, color: 'orange' },
   { key: 24, text: 'Google Gemini', value: 24, color: 'orange' },
+  { key: 28, text: 'Mistral AI', value: 28, color: 'orange' },
   { key: 15, text: '百度文心千帆', value: 15, color: 'blue' },
   { key: 17, text: '阿里通义千问', value: 17, color: 'orange' },
   { key: 18, text: '讯飞星火认知', value: 18, color: 'blue' },

web/default/src/helpers/utils.js

@@ -1,11 +1,13 @@
 import { toast } from 'react-toastify';
 import { toastConstants } from '../constants';
 import React from 'react';
+import { API } from './api';
 
 const HTMLToastContent = ({ htmlContent }) => {
   return <div dangerouslySetInnerHTML={{ __html: htmlContent }} />;
 };
 export default HTMLToastContent;
+
 export function isAdmin() {
   let user = localStorage.getItem('user');
   if (!user) return false;
@@ -29,7 +31,7 @@ export function getSystemName() {
 export function getLogo() {
   let logo = localStorage.getItem('logo');
   if (!logo) return '/logo.png';
-  return logo
+  return logo;
 }
 
 export function getFooterHTML() {
@@ -196,4 +198,30 @@ export function shouldShowPrompt(id) {
 
 export function setPromptShown(id) {
   localStorage.setItem(`prompt-${id}`, 'true');
+}
+
+let channelModels = undefined;
+export async function loadChannelModels() {
+  const res = await API.get('/api/models');
+  const { success, data } = res.data;
+  if (!success) {
+    return;
+  }
+  channelModels = data;
+  localStorage.setItem('channel_models', JSON.stringify(data));
+}
+
+export function getChannelModels(type) {
+  if (channelModels !== undefined && type in channelModels) {
+    return channelModels[type];
+  }
+  let models = localStorage.getItem('channel_models');
+  if (!models) {
+    return [];
+  }
+  channelModels = JSON.parse(models);
+  if (type in channelModels) {
+    return channelModels[type];
+  }
+  return [];
 }

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

@@ -1,7 +1,7 @@
 import React, { useEffect, useState } from 'react';
 import { Button, Form, Header, Input, Message, Segment } from 'semantic-ui-react';
 import { useNavigate, useParams } from 'react-router-dom';
-import { API, showError, showInfo, showSuccess, verifyJSON } from '../../helpers';
+import { API, getChannelModels, showError, showInfo, showSuccess, verifyJSON } from '../../helpers';
 import { CHANNEL_OPTIONS } from '../../constants';
 
 const MODEL_MAPPING_EXAMPLE = {
@@ -56,60 +56,12 @@ const EditChannel = () => {
   const [customModel, setCustomModel] = useState('');
   const handleInputChange = (e, { name, value }) => {
     setInputs((inputs) => ({ ...inputs, [name]: value }));
-    if (name === 'type' && inputs.models.length === 0) {
-      let localModels = [];
-      switch (value) {
-        case 14:
-          localModels = ['claude-instant-1', 'claude-2', 'claude-2.0', 'claude-2.1'];
-          break;
-        case 11:
-          localModels = ['PaLM-2'];
-          break;
-        case 15:
-          localModels = ['ERNIE-Bot', 'ERNIE-Bot-turbo', 'ERNIE-Bot-4', 'Embedding-V1'];
-          break;
-        case 17:
-          localModels = ['qwen-turbo', 'qwen-plus', 'qwen-max', 'qwen-max-longcontext', 'text-embedding-v1'];
-          let withInternetVersion = [];
-          for (let i = 0; i < localModels.length; i++) {
-            if (localModels[i].startsWith('qwen-')) {
-              withInternetVersion.push(localModels[i] + '-internet');
-            }
-          }
-          localModels = [...localModels, ...withInternetVersion];
-          break;
-        case 16:
-          localModels = ["glm-4", "glm-4v", "glm-3-turbo",'chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
-          break;
-        case 18:
-          localModels = [
-            'SparkDesk',
-            'SparkDesk-v1.1',
-            'SparkDesk-v2.1',
-            'SparkDesk-v3.1',
-            'SparkDesk-v3.5'
-          ];
-          break;
-        case 19:
-          localModels = ['360GPT_S2_V9', 'embedding-bert-512-v1', 'embedding_s1_v1', 'semantic_similarity_s1_v1'];
-          break;
-        case 23:
-          localModels = ['hunyuan'];
-          break;
-        case 24:
-          localModels = ['gemini-pro', 'gemini-pro-vision'];
-          break;
-        case 25:
-          localModels = ['moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k'];
-          break;
-        case 26:
-          localModels = ['Baichuan2-Turbo', 'Baichuan2-Turbo-192k', 'Baichuan-Text-Embedding'];
-          break;
-        case 27:
-          localModels = ['abab5.5s-chat', 'abab5.5-chat', 'abab6-chat'];
-          break;
+    if (name === 'type') {
+      let localModels = getChannelModels(value);
+      if (inputs.models.length === 0) {
+        setInputs((inputs) => ({ ...inputs, models: localModels }));
       }
-      setInputs((inputs) => ({ ...inputs, models: localModels }));
+      setBasicModels(localModels);
     }
   };