Merge remote-tracking branch 'remotes/origin_songquanpeng/main'

# Conflicts:
#	relay/adaptor/openai/adaptor.go
#	relay/controller/text.go

relay/controller/audio.go

@@ -7,6 +7,10 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/client"
@@ -21,9 +25,6 @@ import (
 	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
-	"io"
-	"net/http"
-	"strings"
 )
 
 func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
@@ -53,7 +54,7 @@ func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		}
 	}
 
-	modelRatio := billingratio.GetModelRatio(audioModel)
+	modelRatio := billingratio.GetModelRatio(audioModel, channelType)
 	groupRatio := billingratio.GetGroupRatio(group)
 	ratio := modelRatio * groupRatio
 	var quota int64

relay/controller/helper.go

@@ -4,6 +4,10 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"math"
+	"net/http"
+	"strings"
+
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
@@ -16,9 +20,6 @@ import (
 	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
-	"math"
-	"net/http"
-	"strings"
 )
 
 func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.GeneralOpenAIRequest, error) {
@@ -40,78 +41,6 @@ func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.Gener
 	return textRequest, nil
 }
 
-func getImageRequest(c *gin.Context, relayMode int) (*relaymodel.ImageRequest, error) {
-	imageRequest := &relaymodel.ImageRequest{}
-	err := common.UnmarshalBodyReusable(c, imageRequest)
-	if err != nil {
-		return nil, err
-	}
-	if imageRequest.N == 0 {
-		imageRequest.N = 1
-	}
-	if imageRequest.Size == "" {
-		imageRequest.Size = "1024x1024"
-	}
-	if imageRequest.Model == "" {
-		imageRequest.Model = "dall-e-2"
-	}
-	return imageRequest, nil
-}
-
-func isValidImageSize(model string, size string) bool {
-	if model == "cogview-3" {
-		return true
-	}
-	_, ok := billingratio.ImageSizeRatios[model][size]
-	return ok
-}
-
-func getImageSizeRatio(model string, size string) float64 {
-	ratio, ok := billingratio.ImageSizeRatios[model][size]
-	if !ok {
-		return 1
-	}
-	return ratio
-}
-
-func validateImageRequest(imageRequest *relaymodel.ImageRequest, meta *meta.Meta) *relaymodel.ErrorWithStatusCode {
-	// model validation
-	hasValidSize := isValidImageSize(imageRequest.Model, imageRequest.Size)
-	if !hasValidSize {
-		return openai.ErrorWrapper(errors.New("size not supported for this image model"), "size_not_supported", http.StatusBadRequest)
-	}
-	// check prompt length
-	if imageRequest.Prompt == "" {
-		return openai.ErrorWrapper(errors.New("prompt is required"), "prompt_missing", http.StatusBadRequest)
-	}
-	if len(imageRequest.Prompt) > billingratio.ImagePromptLengthLimitations[imageRequest.Model] {
-		return openai.ErrorWrapper(errors.New("prompt is too long"), "prompt_too_long", http.StatusBadRequest)
-	}
-	// Number of generated images validation
-	if !isWithinRange(imageRequest.Model, imageRequest.N) {
-		// channel not azure
-		if meta.ChannelType != channeltype.Azure {
-			return openai.ErrorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
-		}
-	}
-	return nil
-}
-
-func getImageCostRatio(imageRequest *relaymodel.ImageRequest) (float64, error) {
-	if imageRequest == nil {
-		return 0, errors.New("imageRequest is nil")
-	}
-	imageCostRatio := getImageSizeRatio(imageRequest.Model, imageRequest.Size)
-	if imageRequest.Quality == "hd" && imageRequest.Model == "dall-e-3" {
-		if imageRequest.Size == "1024x1024" {
-			imageCostRatio *= 2
-		} else {
-			imageCostRatio *= 1.5
-		}
-	}
-	return imageCostRatio, nil
-}
-
 func getPromptTokens(textRequest *relaymodel.GeneralOpenAIRequest, relayMode int) int {
 	switch relayMode {
 	case relaymode.ChatCompletions:
@@ -167,7 +96,7 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
 		return
 	}
 	var quota int64
-	completionRatio := billingratio.GetCompletionRatio(textRequest.Model)
+	completionRatio := billingratio.GetCompletionRatio(textRequest.Model, meta.ChannelType)
 	promptTokens := usage.PromptTokens
 	completionTokens := usage.CompletionTokens
 	quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))

relay/controller/image.go

@@ -6,7 +6,11 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
+	"net/http"
+
 	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/model"
@@ -16,17 +20,86 @@ import (
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
-	"io"
-	"net/http"
 )
 
-func isWithinRange(element string, value int) bool {
-	if _, ok := billingratio.ImageGenerationAmounts[element]; !ok {
-		return false
+func getImageRequest(c *gin.Context, relayMode int) (*relaymodel.ImageRequest, error) {
+	imageRequest := &relaymodel.ImageRequest{}
+	err := common.UnmarshalBodyReusable(c, imageRequest)
+	if err != nil {
+		return nil, err
 	}
-	min := billingratio.ImageGenerationAmounts[element][0]
-	max := billingratio.ImageGenerationAmounts[element][1]
-	return value >= min && value <= max
+	if imageRequest.N == 0 {
+		imageRequest.N = 1
+	}
+	if imageRequest.Size == "" {
+		imageRequest.Size = "1024x1024"
+	}
+	if imageRequest.Model == "" {
+		imageRequest.Model = "dall-e-2"
+	}
+	return imageRequest, nil
+}
+
+func isValidImageSize(model string, size string) bool {
+	if model == "cogview-3" || billingratio.ImageSizeRatios[model] == nil {
+		return true
+	}
+	_, ok := billingratio.ImageSizeRatios[model][size]
+	return ok
+}
+
+func isValidImagePromptLength(model string, promptLength int) bool {
+	maxPromptLength, ok := billingratio.ImagePromptLengthLimitations[model]
+	return !ok || promptLength <= maxPromptLength
+}
+
+func isWithinRange(element string, value int) bool {
+	amounts, ok := billingratio.ImageGenerationAmounts[element]
+	return !ok || (value >= amounts[0] && value <= amounts[1])
+}
+
+func getImageSizeRatio(model string, size string) float64 {
+	if ratio, ok := billingratio.ImageSizeRatios[model][size]; ok {
+		return ratio
+	}
+	return 1
+}
+
+func validateImageRequest(imageRequest *relaymodel.ImageRequest, meta *meta.Meta) *relaymodel.ErrorWithStatusCode {
+	// check prompt length
+	if imageRequest.Prompt == "" {
+		return openai.ErrorWrapper(errors.New("prompt is required"), "prompt_missing", http.StatusBadRequest)
+	}
+
+	// model validation
+	if !isValidImageSize(imageRequest.Model, imageRequest.Size) {
+		return openai.ErrorWrapper(errors.New("size not supported for this image model"), "size_not_supported", http.StatusBadRequest)
+	}
+
+	if !isValidImagePromptLength(imageRequest.Model, len(imageRequest.Prompt)) {
+		return openai.ErrorWrapper(errors.New("prompt is too long"), "prompt_too_long", http.StatusBadRequest)
+	}
+
+	// Number of generated images validation
+	if !isWithinRange(imageRequest.Model, imageRequest.N) {
+		return openai.ErrorWrapper(errors.New("invalid value of n"), "n_not_within_range", http.StatusBadRequest)
+	}
+	return nil
+}
+
+func getImageCostRatio(imageRequest *relaymodel.ImageRequest) (float64, error) {
+	if imageRequest == nil {
+		return 0, errors.New("imageRequest is nil")
+	}
+	imageCostRatio := getImageSizeRatio(imageRequest.Model, imageRequest.Size)
+	if imageRequest.Quality == "hd" && imageRequest.Model == "dall-e-3" {
+		if imageRequest.Size == "1024x1024" {
+			imageCostRatio *= 2
+		} else {
+			imageCostRatio *= 1.5
+		}
+	}
+	return imageCostRatio, nil
 }
 
 func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
@@ -95,7 +168,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		requestBody = bytes.NewBuffer(jsonStr)
 	}
 
-	modelRatio := billingratio.GetModelRatio(imageModel)
+	modelRatio := billingratio.GetModelRatio(imageModel, meta.ChannelType)
 	groupRatio := billingratio.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
 	userQuota, err := model.CacheGetUserQuota(ctx, meta.UserId)

relay/controller/proxy.go

@@ -0,0 +1,41 @@
+// Package controller is a package for handling the relay controller
+package controller
+
+import (
+	"fmt"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/relay"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/meta"
+	relaymodel "github.com/songquanpeng/one-api/relay/model"
+)
+
+// RelayProxyHelper is a helper function to proxy the request to the upstream service
+func RelayProxyHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
+	ctx := c.Request.Context()
+	meta := meta.GetByContext(c)
+
+	adaptor := relay.GetAdaptor(meta.APIType)
+	if adaptor == nil {
+		return openai.ErrorWrapper(fmt.Errorf("invalid api type: %d", meta.APIType), "invalid_api_type", http.StatusBadRequest)
+	}
+	adaptor.Init(meta)
+
+	resp, err := adaptor.DoRequest(c, meta, c.Request.Body)
+	if err != nil {
+		logger.Errorf(ctx, "DoRequest failed: %s", err.Error())
+		return openai.ErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
+	}
+
+	// do response
+	_, respErr := adaptor.DoResponse(c, resp, meta)
+	if respErr != nil {
+		logger.Errorf(ctx, "respErr is not nil: %+v", respErr)
+		return respErr
+	}
+
+	return nil
+}

relay/controller/text.go

@@ -4,6 +4,10 @@ import (
 	"bytes"
 	"encoding/json"
 	"fmt"
+	"github.com/songquanpeng/one-api/relay/adaptor"
+	"io"
+	"net/http"
+
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay"
@@ -14,15 +18,8 @@ import (
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
-	"io"
-	"math"
-	"net/http"
-	"os"
-	"strings"
 )
 
-var fixAIProxyGpt4oMaxTokens = os.Getenv("FIX_AI_PROXY_GPT4O_MAX_TOKENS") == "1"
-
 func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	ctx := c.Request.Context()
 	meta := meta.GetByContext(c)
@@ -35,12 +32,11 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	meta.IsStream = textRequest.Stream
 
 	// map model name
-	var isModelMapped bool
 	meta.OriginModelName = textRequest.Model
-	textRequest.Model, isModelMapped = getMappedModelName(textRequest.Model, meta.ModelMapping)
+	textRequest.Model, _ = getMappedModelName(textRequest.Model, meta.ModelMapping)
 	meta.ActualModelName = textRequest.Model
 	// get model ratio & group ratio
-	modelRatio := billingratio.GetModelRatio(textRequest.Model)
+	modelRatio := billingratio.GetModelRatio(textRequest.Model, meta.ChannelType)
 	groupRatio := billingratio.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
 	// pre-consume quota
@@ -58,41 +54,9 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	}
 	adaptor.Init(meta)
 
-	// get request body
-	var requestBody io.Reader
-	if meta.APIType == apitype.OpenAI {
-		// no need to convert request for openai
-		shouldResetRequestBody := isModelMapped ||
-			meta.ChannelType == channeltype.Baichuan /*frequency_penalty 0 is not acceptable for baichuan*/ ||
-			(meta.ChannelType == channeltype.AIProxy && fixAIProxyGpt4oMaxTokens && strings.HasPrefix(textRequest.Model, "gpt-4o"))
-		if shouldResetRequestBody {
-			if meta.ChannelType == channeltype.AIProxy {
-				maxTokens := textRequest.MaxTokens
-				maxTokens = int(math.Min(float64(maxTokens), 4096))
-				if maxTokens == 0 {
-					maxTokens = 4096
-				}
-				textRequest.MaxTokens = maxTokens
-			}
-			jsonStr, err := json.Marshal(textRequest)
-			if err != nil {
-				return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
-			}
-			requestBody = bytes.NewBuffer(jsonStr)
-		} else {
-			requestBody = c.Request.Body
-		}
-	} else {
-		convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
-		if err != nil {
-			return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
-		}
-		jsonData, err := json.Marshal(convertedRequest)
-		if err != nil {
-			return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
-		}
-		logger.Debugf(ctx, "converted request: \n%s", string(jsonData))
-		requestBody = bytes.NewBuffer(jsonData)
+	requestBody, err := getRequestBody(c, meta, textRequest, adaptor)
+	if err != nil {
+		return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
 	}
 
 	// do request
@@ -117,3 +81,26 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	go postConsumeQuota(ctx, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio)
 	return nil
 }
+
+func getRequestBody(c *gin.Context, meta *meta.Meta, textRequest *model.GeneralOpenAIRequest, adaptor adaptor.Adaptor) (io.Reader, error) {
+	if meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
+		// no need to convert request for openai
+		return c.Request.Body, nil
+	}
+
+	// get request body
+	var requestBody io.Reader
+	convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
+	if err != nil {
+		logger.Debugf(c.Request.Context(), "converted request failed: %s\n", err.Error())
+		return nil, err
+	}
+	jsonData, err := json.Marshal(convertedRequest)
+	if err != nil {
+		logger.Debugf(c.Request.Context(), "converted request json_marshal_failed: %s\n", err.Error())
+		return nil, err
+	}
+	logger.Debugf(c.Request.Context(), "converted request: \n%s", string(jsonData))
+	requestBody = bytes.NewBuffer(jsonData)
+	return requestBody, nil
+}