Merge branch 'main' into patch/images-edits

relay/adaptor/minimax/constants.go

@@ -1,7 +1,11 @@
 package minimax
 
+// https://www.minimaxi.com/document/guides/chat-model/V2?id=65e0736ab2845de20908e2dd
+
 var ModelList = []string{
-	"abab5.5s-chat",
-	"abab5.5-chat",
+	"abab6.5-chat",
+	"abab6.5s-chat",
 	"abab6-chat",
+	"abab5.5-chat",
+	"abab5.5s-chat",
 }

relay/adaptor/openai/adaptor.go

@@ -86,9 +86,13 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Met
 	if meta.IsStream {
 		var responseText string
 		err, responseText, usage = StreamHandler(c, resp, meta.Mode)
-		if usage == nil {
+		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:

relay/adaptor/openai/compatible.go

@@ -10,6 +10,7 @@ import (
 	"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"
 )
 
@@ -24,6 +25,7 @@ var CompatibleChannels = []int{
 	channeltype.LingYiWanWu,
 	channeltype.StepFun,
 	channeltype.DeepSeek,
+	channeltype.TogetherAI,
 }
 
 func GetCompatibleChannelMeta(channelType int) (string, []string) {
@@ -48,6 +50,8 @@ func GetCompatibleChannelMeta(channelType int) (string, []string) {
 		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

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

relay/adaptor/openai/token.go

@@ -24,6 +24,10 @@ func InitTokenEncoders() {
 		logger.FatalLog(fmt.Sprintf("failed to get gpt-3.5-turbo token encoder: %s", err.Error()))
 	}
 	defaultTokenEncoder = gpt35TokenEncoder
+	gpt4oTokenEncoder, err := tiktoken.EncodingForModel("gpt-4o")
+	if err != nil {
+		logger.FatalLog(fmt.Sprintf("failed to get gpt-4o token encoder: %s", err.Error()))
+	}
 	gpt4TokenEncoder, err := tiktoken.EncodingForModel("gpt-4")
 	if err != nil {
 		logger.FatalLog(fmt.Sprintf("failed to get gpt-4 token encoder: %s", err.Error()))
@@ -31,6 +35,8 @@ func InitTokenEncoders() {
 	for model := range billingratio.ModelRatio {
 		if strings.HasPrefix(model, "gpt-3.5") {
 			tokenEncoderMap[model] = gpt35TokenEncoder
+		} else if strings.HasPrefix(model, "gpt-4o") {
+			tokenEncoderMap[model] = gpt4oTokenEncoder
 		} else if strings.HasPrefix(model, "gpt-4") {
 			tokenEncoderMap[model] = gpt4TokenEncoder
 		} else {

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/billing/ratio/image.go

@@ -49,3 +49,8 @@ var ImagePromptLengthLimitations = map[string]int{
 	"wanx-v1":                   4000,
 	"cogview-3":                 833,
 }
+
+var ImageOriginModelName = map[string]string{
+	"ali-stable-diffusion-xl":   "stable-diffusion-xl",
+	"ali-stable-diffusion-v1.5": "stable-diffusion-v1.5",
+}

relay/billing/ratio/model.go

@@ -32,6 +32,8 @@ var ModelRatio = map[string]float64{
 	"gpt-4-turbo-preview":     5,    // $0.01 / 1K tokens
 	"gpt-4-turbo":             5,    // $0.01 / 1K tokens
 	"gpt-4-turbo-2024-04-09":  5,    // $0.01 / 1K tokens
+	"gpt-4o":                  2.5,  // $0.005 / 1K tokens
+	"gpt-4o-2024-05-13":       2.5,  // $0.005 / 1K tokens
 	"gpt-4-vision-preview":    5,    // $0.01 / 1K tokens
 	"gpt-3.5-turbo":           0.25, // $0.0005 / 1K tokens
 	"gpt-3.5-turbo-0301":      0.75,
@@ -138,6 +140,8 @@ var ModelRatio = map[string]float64{
 	"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,
@@ -169,7 +173,7 @@ var ModelRatio = map[string]float64{
 	"command-light":         0.5,
 	"command-light-nightly": 0.5,
 	"command-r":             0.5 / 1000 * USD,
-	"command-r-plus	":       3.0 / 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,
@@ -278,7 +282,9 @@ func GetCompletionRatio(name string) float64 {
 		return 4.0 / 3.0
 	}
 	if strings.HasPrefix(name, "gpt-4") {
-		if strings.HasPrefix(name, "gpt-4-turbo") || strings.HasSuffix(name, "preview") {
+		if strings.HasPrefix(name, "gpt-4-turbo") ||
+			strings.HasPrefix(name, "gpt-4o") ||
+			strings.HasSuffix(name, "preview") {
 			return 3
 		}
 		return 2

relay/channeltype/define.go

@@ -40,6 +40,7 @@ const (
 	DeepSeek
 	Cloudflare
 	DeepL
+	TogetherAI
 
 	Dummy
 )

relay/channeltype/url.go

@@ -40,6 +40,7 @@ var ChannelBaseURLs = []string{
 	"https://api.deepseek.com",                  // 36
 	"https://api.cloudflare.com",                // 37
 	"https://api-free.deepl.com",                // 38
+	"https://api.together.xyz",                  // 39
 }
 
 func init() {

relay/controller/error.go

@@ -53,6 +53,16 @@ func (e GeneralErrorResponse) ToMessage() string {
 }
 
 func RelayErrorHandler(resp *http.Response) (ErrorWithStatusCode *model.ErrorWithStatusCode) {
+	if resp == nil {
+		return &model.ErrorWithStatusCode{
+			StatusCode: 500,
+			Error: model.Error{
+				Message: "resp is nil",
+				Type:    "upstream_error",
+				Code:    "bad_response",
+			},
+		}
+	}
 	ErrorWithStatusCode = &model.ErrorWithStatusCode{
 		StatusCode: resp.StatusCode,
 		Error: model.Error{

relay/controller/helper.go

@@ -208,6 +208,9 @@ func getMappedModelName(modelName string, mapping map[string]string) (string, bo
 
 func isErrorHappened(meta *meta.Meta, resp *http.Response) bool {
 	if resp == nil {
+		if meta.ChannelType == channeltype.AwsClaude {
+			return false
+		}
 		return true
 	}
 	if resp.StatusCode != http.StatusOK {

relay/controller/image.go

@@ -57,6 +57,11 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		return openai.ErrorWrapper(err, "get_image_cost_ratio_failed", http.StatusInternalServerError)
 	}
 
+	imageModel := imageRequest.Model
+	// Convert the original image model
+	imageRequest.Model, _ = getMappedModelName(imageRequest.Model, billingratio.ImageOriginModelName)
+	c.Set("response_format", imageRequest.ResponseFormat)
+
 	var requestBody io.Reader
 	if strings.ToLower(c.GetString(ctxkey.ContentType)) == "application/json" &&
 		isModelMapped || meta.ChannelType == channeltype.Azure { // make Azure channel request body
@@ -92,7 +97,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		requestBody = bytes.NewBuffer(jsonStr)
 	}
 
-	modelRatio := billingratio.GetModelRatio(imageRequest.Model)
+	modelRatio := billingratio.GetModelRatio(imageModel)
 	groupRatio := billingratio.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
 	userQuota, err := model.CacheGetUserQuota(ctx, meta.UserId)