oadmin/new-api
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor: Simplify model mapping and pricing logic across relay modules

Browse Source
This commit is contained in:
1808837298@qq.com
2025-02-20 16:41:46 +08:00
parent 60aac77c08
commit 06da65a9d0
13 changed files with 279 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files

67
relay/common/relay_info.go
View File

@@ -13,24 +13,24 @@ import (
)
type RelayInfo struct {
ChannelType int
ChannelId int
TokenId int
TokenKey string
UserId int
Group string
TokenUnlimited bool
StartTime time.Time
FirstResponseTime time.Time
setFirstResponse bool
ApiType int
IsStream bool
IsPlayground bool
UsePrice bool
RelayMode int
UpstreamModelName string
OriginModelName string
RecodeModelName string
ChannelType int
ChannelId int
TokenId int
TokenKey string
UserId int
Group string
TokenUnlimited bool
StartTime time.Time
FirstResponseTime time.Time
setFirstResponse bool
ApiType int
IsStream bool
IsPlayground bool
UsePrice bool
RelayMode int
UpstreamModelName string
OriginModelName string
//RecodeModelName string
RequestURLPath string
ApiVersion string
PromptTokens int
@@ -39,6 +39,7 @@ type RelayInfo struct {
BaseUrl string
SupportStreamOptions bool
ShouldIncludeUsage bool
IsModelMapped bool
ClientWs *websocket.Conn
TargetWs *websocket.Conn
InputAudioFormat string
@@ -50,6 +51,18 @@ type RelayInfo struct {
ChannelSetting map[string]interface{}
}
// 定义支持流式选项的通道类型
var streamSupportedChannels = map[int]bool{
common.ChannelTypeOpenAI: true,
common.ChannelTypeAnthropic: true,
common.ChannelTypeAws: true,
common.ChannelTypeGemini: true,
common.ChannelCloudflare: true,
common.ChannelTypeAzure: true,
common.ChannelTypeVolcEngine: true,
common.ChannelTypeOllama: true,
}
func GenRelayInfoWs(c *gin.Context, ws *websocket.Conn) *RelayInfo {
info := GenRelayInfo(c)
info.ClientWs = ws
@@ -89,12 +102,13 @@ func GenRelayInfo(c *gin.Context) *RelayInfo {
FirstResponseTime: startTime.Add(-time.Second),
OriginModelName: c.GetString("original_model"),
UpstreamModelName: c.GetString("original_model"),
RecodeModelName: c.GetString("recode_model"),
ApiType: apiType,
ApiVersion: c.GetString("api_version"),
ApiKey: strings.TrimPrefix(c.Request.Header.Get("Authorization"), "Bearer "),
Organization: c.GetString("channel_organization"),
ChannelSetting: channelSetting,
//RecodeModelName: c.GetString("original_model"),
IsModelMapped: false,
ApiType: apiType,
ApiVersion: c.GetString("api_version"),
ApiKey: strings.TrimPrefix(c.Request.Header.Get("Authorization"), "Bearer "),
Organization: c.GetString("channel_organization"),
ChannelSetting: channelSetting,
}
if strings.HasPrefix(c.Request.URL.Path, "/pg") {
info.IsPlayground = true
@@ -110,10 +124,7 @@ func GenRelayInfo(c *gin.Context) *RelayInfo {
if info.ChannelType == common.ChannelTypeVertexAi {
info.ApiVersion = c.GetString("region")
}
if info.ChannelType == common.ChannelTypeOpenAI || info.ChannelType == common.ChannelTypeAnthropic ||
info.ChannelType == common.ChannelTypeAws || info.ChannelType == common.ChannelTypeGemini ||
info.ChannelType == common.ChannelCloudflare || info.ChannelType == common.ChannelTypeAzure ||
info.ChannelType == common.ChannelTypeVolcEngine || info.ChannelType == common.ChannelTypeOllama {
if streamSupportedChannels[info.ChannelType] {
info.SupportStreamOptions = true
}
return info

25
relay/helper/model_mapped.go Normal file
View File

@@ -0,0 +1,25 @@
package helper
import (
"encoding/json"
"fmt"
"github.com/gin-gonic/gin"
"one-api/relay/common"
)
func ModelMappedHelper(c *gin.Context, info *common.RelayInfo) error {
// map model name
modelMapping := c.GetString("model_mapping")
if modelMapping != "" && modelMapping != "{}" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return fmt.Errorf("unmarshal_model_mapping_failed")
}
if modelMap[info.OriginModelName] != "" {
info.UpstreamModelName = modelMap[info.OriginModelName]
info.IsModelMapped = true
}
}
return nil
}

41
relay/helper/price.go Normal file
View File

@@ -0,0 +1,41 @@
package helper
import (
"github.com/gin-gonic/gin"
"one-api/common"
relaycommon "one-api/relay/common"
"one-api/setting"
)
type PriceData struct {
ModelPrice float64
ModelRatio float64
GroupRatio float64
UsePrice bool
ShouldPreConsumedQuota int
}
func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens int, maxTokens int) PriceData {
modelPrice, usePrice := common.GetModelPrice(info.OriginModelName, false)
groupRatio := setting.GetGroupRatio(info.Group)
var preConsumedQuota int
var modelRatio float64
if !usePrice {
preConsumedTokens := common.PreConsumedQuota
if maxTokens != 0 {
preConsumedTokens = promptTokens + maxTokens
}
modelRatio = common.GetModelRatio(info.OriginModelName)
ratio := modelRatio * groupRatio
preConsumedQuota = int(float64(preConsumedTokens) * ratio)
} else {
preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
}
return PriceData{
ModelPrice: modelPrice,
ModelRatio: modelRatio,
GroupRatio: groupRatio,
UsePrice: usePrice,
ShouldPreConsumedQuota: preConsumedQuota,
}
}

29
relay/relay-audio.go
View File

@@ -1,7 +1,6 @@
package relay
import (
"encoding/json"
"errors"
"fmt"
"github.com/gin-gonic/gin"
@@ -11,6 +10,7 @@ import (
"one-api/model"
relaycommon "one-api/relay/common"
relayconstant "one-api/relay/constant"
"one-api/relay/helper"
"one-api/service"
"one-api/setting"
)
@@ -73,15 +73,13 @@ func AudioHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
relayInfo.PromptTokens = promptTokens
}
modelRatio := common.GetModelRatio(audioRequest.Model)
groupRatio := setting.GetGroupRatio(relayInfo.Group)
ratio := modelRatio * groupRatio
preConsumedQuota := int(float64(preConsumedTokens) * ratio)
priceData := helper.ModelPriceHelper(c, relayInfo, preConsumedTokens, 0)
userQuota, err := model.GetUserQuota(relayInfo.UserId, false)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "get_user_quota_failed", http.StatusInternalServerError)
}
preConsumedQuota, userQuota, openaiErr = preConsumeQuota(c, preConsumedQuota, relayInfo)
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
if openaiErr != nil {
return openaiErr
}
@@ -91,19 +89,12 @@ func AudioHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
}
}()
// map model name
modelMapping := c.GetString("model_mapping")
if modelMapping != "" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return service.OpenAIErrorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
}
if modelMap[audioRequest.Model] != "" {
audioRequest.Model = modelMap[audioRequest.Model]
}
err = helper.ModelMappedHelper(c, relayInfo)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusInternalServerError)
}
relayInfo.UpstreamModelName = audioRequest.Model
audioRequest.Model = relayInfo.UpstreamModelName
adaptor := GetAdaptor(relayInfo.ApiType)
if adaptor == nil {
@@ -140,7 +131,7 @@ func AudioHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
return openaiErr
}
postConsumeQuota(c, relayInfo, audioRequest.Model, usage.(*dto.Usage), ratio, preConsumedQuota, userQuota, modelRatio, groupRatio, 0, false, "")
postConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
return nil
}

31
relay/relay-image.go
View File

@@ -12,6 +12,7 @@ import (
"one-api/dto"
"one-api/model"
relaycommon "one-api/relay/common"
"one-api/relay/helper"
"one-api/service"
"one-api/setting"
"strings"
@@ -68,7 +69,7 @@ func getAndValidImageRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.
return imageRequest, nil
}
func ImageHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
func ImageHelper(c *gin.Context) *dto.OpenAIErrorWithStatusCode {
relayInfo := relaycommon.GenRelayInfo(c)
imageRequest, err := getAndValidImageRequest(c, relayInfo)
@@ -77,19 +78,12 @@ func ImageHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
return service.OpenAIErrorWrapper(err, "invalid_image_request", http.StatusBadRequest)
}
// map model name
modelMapping := c.GetString("model_mapping")
if modelMapping != "" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return service.OpenAIErrorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
}
if modelMap[imageRequest.Model] != "" {
imageRequest.Model = modelMap[imageRequest.Model]
}
err = helper.ModelMappedHelper(c, relayInfo)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusInternalServerError)
}
relayInfo.UpstreamModelName = imageRequest.Model
imageRequest.Model = relayInfo.UpstreamModelName
modelPrice, success := common.GetModelPrice(imageRequest.Model, true)
if !success {
@@ -183,8 +177,15 @@ func ImageHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
quality = "hd"
}
logContent := fmt.Sprintf("大小 %s, 品质 %s", imageRequest.Size, quality)
postConsumeQuota(c, relayInfo, imageRequest.Model, usage, 0, 0, userQuota, 0, groupRatio, imageRatio, true, logContent)
priceData := helper.PriceData{
UsePrice: true,
GroupRatio: groupRatio,
ModelPrice: modelPrice,
ModelRatio: 0,
ShouldPreConsumedQuota: 0,
}
logContent := fmt.Sprintf("大小 %s, 品质 %s", imageRequest.Size, quality)
postConsumeQuota(c, relayInfo, usage, 0, userQuota, priceData, logContent)
return nil
}

68
relay/relay-text.go
View File

@@ -15,6 +15,7 @@ import (
"one-api/model"
relaycommon "one-api/relay/common"
relayconstant "one-api/relay/constant"
"one-api/relay/helper"
"one-api/service"
"one-api/setting"
"strings"
@@ -76,33 +77,6 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
return service.OpenAIErrorWrapperLocal(err, "invalid_text_request", http.StatusBadRequest)
}
// map model name
//isModelMapped := false
modelMapping := c.GetString("model_mapping")
//isModelMapped := false
if modelMapping != "" && modelMapping != "{}" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
}
if modelMap[textRequest.Model] != "" {
//isModelMapped = true
textRequest.Model = modelMap[textRequest.Model]
// set upstream model name
//isModelMapped = true
}
}
relayInfo.UpstreamModelName = textRequest.Model
relayInfo.RecodeModelName = textRequest.Model
modelPrice, getModelPriceSuccess := common.GetModelPrice(textRequest.Model, false)
groupRatio := setting.GetGroupRatio(relayInfo.Group)
var preConsumedQuota int
var ratio float64
var modelRatio float64
//err := service.SensitiveWordsCheck(textRequest)
if setting.ShouldCheckPromptSensitive() {
err = checkRequestSensitive(textRequest, relayInfo)
if err != nil {
@@ -110,6 +84,13 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
}
}
err = helper.ModelMappedHelper(c, relayInfo)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusInternalServerError)
}
textRequest.Model = relayInfo.UpstreamModelName
// 获取 promptTokens，如果上下文中已经存在，则直接使用
var promptTokens int
if value, exists := c.Get("prompt_tokens"); exists {
@@ -124,20 +105,10 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
c.Set("prompt_tokens", promptTokens)
}
if !getModelPriceSuccess {
preConsumedTokens := common.PreConsumedQuota
if textRequest.MaxTokens != 0 {
preConsumedTokens = promptTokens + int(textRequest.MaxTokens)
}
modelRatio = common.GetModelRatio(textRequest.Model)
ratio = modelRatio * groupRatio
preConsumedQuota = int(float64(preConsumedTokens) * ratio)
} else {
preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
}
priceData := helper.ModelPriceHelper(c, relayInfo, promptTokens, int(textRequest.MaxTokens))
// pre-consume quota 预消耗配额
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, preConsumedQuota, relayInfo)
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
if openaiErr != nil {
return openaiErr
}
@@ -220,10 +191,10 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
return openaiErr
}
if strings.HasPrefix(relayInfo.RecodeModelName, "gpt-4o-audio") {
service.PostAudioConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, modelRatio, groupRatio, modelPrice, getModelPriceSuccess, "")
if strings.HasPrefix(relayInfo.OriginModelName, "gpt-4o-audio") {
service.PostAudioConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
} else {
postConsumeQuota(c, relayInfo, relayInfo.RecodeModelName, usage.(*dto.Usage), ratio, preConsumedQuota, userQuota, modelRatio, groupRatio, modelPrice, getModelPriceSuccess, "")
postConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
}
return nil
}
@@ -319,9 +290,8 @@ func returnPreConsumedQuota(c *gin.Context, relayInfo *relaycommon.RelayInfo, us
}
}
func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelName string,
usage *dto.Usage, ratio float64, preConsumedQuota int, userQuota int, modelRatio float64, groupRatio float64,
modelPrice float64, usePrice bool, extraContent string) {
func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
usage *dto.Usage, preConsumedQuota int, userQuota int, priceData helper.PriceData, extraContent string) {
if usage == nil {
usage = &dto.Usage{
PromptTokens: relayInfo.PromptTokens,
@@ -333,12 +303,18 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelN
useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
promptTokens := usage.PromptTokens
completionTokens := usage.CompletionTokens
modelName := relayInfo.OriginModelName
tokenName := ctx.GetString("token_name")
completionRatio := common.GetCompletionRatio(modelName)
ratio := priceData.ModelRatio * priceData.GroupRatio
modelRatio := priceData.ModelRatio
groupRatio := priceData.GroupRatio
modelPrice := priceData.ModelPrice
usePrice := priceData.UsePrice
quota := 0
if !usePrice {
if !priceData.UsePrice {
quota = promptTokens + int(math.Round(float64(completionTokens)*completionRatio))
quota = int(math.Round(float64(quota) * ratio))
if ratio != 0 && quota <= 0 {

41
relay/relay_embedding.go
View File

@@ -10,8 +10,8 @@ import (
"one-api/dto"
relaycommon "one-api/relay/common"
relayconstant "one-api/relay/constant"
"one-api/relay/helper"
"one-api/service"
"one-api/setting"
)
func getEmbeddingPromptToken(embeddingRequest dto.EmbeddingRequest) int {
@@ -47,43 +47,20 @@ func EmbeddingHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode)
return service.OpenAIErrorWrapperLocal(err, "invalid_embedding_request", http.StatusBadRequest)
}
// map model name
modelMapping := c.GetString("model_mapping")
//isModelMapped := false
if modelMapping != "" && modelMapping != "{}" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
}
if modelMap[embeddingRequest.Model] != "" {
embeddingRequest.Model = modelMap[embeddingRequest.Model]
// set upstream model name
//isModelMapped = true
}
err = helper.ModelMappedHelper(c, relayInfo)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusInternalServerError)
}
relayInfo.UpstreamModelName = embeddingRequest.Model
modelPrice, success := common.GetModelPrice(embeddingRequest.Model, false)
groupRatio := setting.GetGroupRatio(relayInfo.Group)
var preConsumedQuota int
var ratio float64
var modelRatio float64
embeddingRequest.Model = relayInfo.UpstreamModelName
promptToken := getEmbeddingPromptToken(*embeddingRequest)
if !success {
preConsumedTokens := promptToken
modelRatio = common.GetModelRatio(embeddingRequest.Model)
ratio = modelRatio * groupRatio
preConsumedQuota = int(float64(preConsumedTokens) * ratio)
} else {
preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
}
relayInfo.PromptTokens = promptToken
priceData := helper.ModelPriceHelper(c, relayInfo, promptToken, 0)
// pre-consume quota 预消耗配额
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, preConsumedQuota, relayInfo)
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
if openaiErr != nil {
return openaiErr
}
@@ -132,6 +109,6 @@ func EmbeddingHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode)
service.ResetStatusCode(openaiErr, statusCodeMappingStr)
return openaiErr
}
postConsumeQuota(c, relayInfo, embeddingRequest.Model, usage.(*dto.Usage), ratio, preConsumedQuota, userQuota, modelRatio, groupRatio, modelPrice, success, "")
postConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
return nil
}

41
relay/relay_rerank.go
View File

@@ -9,8 +9,8 @@ import (
"one-api/common"
"one-api/dto"
relaycommon "one-api/relay/common"
"one-api/relay/helper"
"one-api/service"
"one-api/setting"
)
func getRerankPromptToken(rerankRequest dto.RerankRequest) int {
@@ -40,43 +40,20 @@ func RerankHelper(c *gin.Context, relayMode int) (openaiErr *dto.OpenAIErrorWith
return service.OpenAIErrorWrapperLocal(fmt.Errorf("documents is empty"), "invalid_documents", http.StatusBadRequest)
}
// map model name
modelMapping := c.GetString("model_mapping")
//isModelMapped := false
if modelMapping != "" && modelMapping != "{}" {
modelMap := make(map[string]string)
err := json.Unmarshal([]byte(modelMapping), &modelMap)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
}
if modelMap[rerankRequest.Model] != "" {
rerankRequest.Model = modelMap[rerankRequest.Model]
// set upstream model name
//isModelMapped = true
}
err = helper.ModelMappedHelper(c, relayInfo)
if err != nil {
return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusInternalServerError)
}
relayInfo.UpstreamModelName = rerankRequest.Model
modelPrice, success := common.GetModelPrice(rerankRequest.Model, false)
groupRatio := setting.GetGroupRatio(relayInfo.Group)
var preConsumedQuota int
var ratio float64
var modelRatio float64
rerankRequest.Model = relayInfo.UpstreamModelName
promptToken := getRerankPromptToken(*rerankRequest)
if !success {
preConsumedTokens := promptToken
modelRatio = common.GetModelRatio(rerankRequest.Model)
ratio = modelRatio * groupRatio
preConsumedQuota = int(float64(preConsumedTokens) * ratio)
} else {
preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
}
relayInfo.PromptTokens = promptToken
priceData := helper.ModelPriceHelper(c, relayInfo, promptToken, 0)
// pre-consume quota 预消耗配额
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, preConsumedQuota, relayInfo)
preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
if openaiErr != nil {
return openaiErr
}
@@ -124,6 +101,6 @@ func RerankHelper(c *gin.Context, relayMode int) (openaiErr *dto.OpenAIErrorWith
service.ResetStatusCode(openaiErr, statusCodeMappingStr)
return openaiErr
}
postConsumeQuota(c, relayInfo, rerankRequest.Model, usage.(*dto.Usage), ratio, preConsumedQuota, userQuota, modelRatio, groupRatio, modelPrice, success, "")
postConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
return nil
}