xinyin025/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-11-12 03:13:41 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Refactor codebase, introduce relaymode package, update constants and improve consistency

Browse Source 
- Refactor constant definitions and organization
- Clean up package level variables and functions
- Introduce new `relaymode` and `apitype` packages for constant definitions
- Refactor and simplify code in several packages including `openai`, `relay/channel/baidu`, `relay/util`, `relay/controller`, `relay/channeltype`
- Add helper functions in `relay/channeltype` package to convert channel type constants to corresponding API type constants
- Remove deprecated functions such as `ResponseText2Usage` from `relay/channel/openai/helper.go`
- Modify code in `relay/util/validation.go` and related files to use new `validator.ValidateTextRequest` function
- Rename `util` package to `relaymode` and update related imports in several packages
This commit is contained in:
Laisky.Cai
2024-04-06 05:18:04 +00:00
parent 7c59a97a6b
commit 50bab08496
157 changed files with 3217 additions and 19160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
relay/controller/helper.go
View File

@@ -9,10 +9,13 @@ import (
"github.com/songquanpeng/one-api/common/config"
"github.com/songquanpeng/one-api/common/logger"
"github.com/songquanpeng/one-api/model"
"github.com/songquanpeng/one-api/relay/channel/openai"
"github.com/songquanpeng/one-api/relay/constant"
"github.com/songquanpeng/one-api/relay/adaptor/openai"
billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
"github.com/songquanpeng/one-api/relay/channeltype"
"github.com/songquanpeng/one-api/relay/controller/validator"
"github.com/songquanpeng/one-api/relay/meta"
relaymodel "github.com/songquanpeng/one-api/relay/model"
"github.com/songquanpeng/one-api/relay/util"
"github.com/songquanpeng/one-api/relay/relaymode"
"math"
"net/http"
)
@@ -23,21 +26,21 @@ func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.Gener
if err != nil {
return nil, err
}
if relayMode == constant.RelayModeModerations && textRequest.Model == "" {
if relayMode == relaymode.Moderations && textRequest.Model == "" {
textRequest.Model = "text-moderation-latest"
}
if relayMode == constant.RelayModeEmbeddings && textRequest.Model == "" {
if relayMode == relaymode.Embeddings && textRequest.Model == "" {
textRequest.Model = c.Param("model")
}
err = util.ValidateTextRequest(textRequest, relayMode)
err = validator.ValidateTextRequest(textRequest, relayMode)
if err != nil {
return nil, err
}
return textRequest, nil
}
func getImageRequest(c *gin.Context, relayMode int) (*openai.ImageRequest, error) {
imageRequest := &openai.ImageRequest{}
func getImageRequest(c *gin.Context, relayMode int) (*relaymodel.ImageRequest, error) {
imageRequest := &relaymodel.ImageRequest{}
err := common.UnmarshalBodyReusable(c, imageRequest)
if err != nil {
return nil, err
@@ -54,9 +57,25 @@ func getImageRequest(c *gin.Context, relayMode int) (*openai.ImageRequest, error
return imageRequest, nil
}
func validateImageRequest(imageRequest *openai.ImageRequest, meta *util.RelayMeta) *relaymodel.ErrorWithStatusCode {
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 := constant.DalleSizeRatios[imageRequest.Model][imageRequest.Size]
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)
}
@@ -64,27 +83,24 @@ func validateImageRequest(imageRequest *openai.ImageRequest, meta *util.RelayMet
if imageRequest.Prompt == "" {
return openai.ErrorWrapper(errors.New("prompt is required"), "prompt_missing", http.StatusBadRequest)
}
if len(imageRequest.Prompt) > constant.DalleImagePromptLengthLimitations[imageRequest.Model] {
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 != common.ChannelTypeAzure {
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 *openai.ImageRequest) (float64, error) {
func getImageCostRatio(imageRequest *relaymodel.ImageRequest) (float64, error) {
if imageRequest == nil {
return 0, errors.New("imageRequest is nil")
}
imageCostRatio, hasValidSize := constant.DalleSizeRatios[imageRequest.Model][imageRequest.Size]
if !hasValidSize {
return 0, errors.Errorf("size not supported for this image model: %s", imageRequest.Size)
}
imageCostRatio := getImageSizeRatio(imageRequest.Model, imageRequest.Size)
if imageRequest.Quality == "hd" && imageRequest.Model == "dall-e-3" {
if imageRequest.Size == "1024x1024" {
imageCostRatio *= 2
@@ -97,11 +113,11 @@ func getImageCostRatio(imageRequest *openai.ImageRequest) (float64, error) {
func getPromptTokens(textRequest *relaymodel.GeneralOpenAIRequest, relayMode int) int {
switch relayMode {
case constant.RelayModeChatCompletions:
case relaymode.ChatCompletions:
return openai.CountTokenMessages(textRequest.Messages, textRequest.Model)
case constant.RelayModeCompletions:
case relaymode.Completions:
return openai.CountTokenInput(textRequest.Prompt, textRequest.Model)
case constant.RelayModeModerations:
case relaymode.Moderations:
return openai.CountTokenInput(textRequest.Input, textRequest.Model)
}
return 0
@@ -115,7 +131,7 @@ func getPreConsumedQuota(textRequest *relaymodel.GeneralOpenAIRequest, promptTok
return int64(float64(preConsumedTokens) * ratio)
}
func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIRequest, promptTokens int, ratio float64, meta *util.RelayMeta) (int64, *relaymodel.ErrorWithStatusCode) {
func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIRequest, promptTokens int, ratio float64, meta *meta.Meta) (int64, *relaymodel.ErrorWithStatusCode) {
preConsumedQuota := getPreConsumedQuota(textRequest, promptTokens, ratio)
userQuota, err := model.CacheGetUserQuota(ctx, meta.UserId)
@@ -144,13 +160,13 @@ func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIR
return preConsumedQuota, nil
}
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *util.RelayMeta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64) {
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64) {
if usage == nil {
logger.Error(ctx, "usage is nil, which is unexpected")
return
}
var quota int64
completionRatio := common.GetCompletionRatio(textRequest.Model)
completionRatio := billingratio.GetCompletionRatio(textRequest.Model)
promptTokens := usage.PromptTokens
completionTokens := usage.CompletionTokens
quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
@@ -178,3 +194,14 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *util.R
model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
model.UpdateChannelUsedQuota(meta.ChannelId, quota)
}
func getMappedModelName(modelName string, mapping map[string]string) (string, bool) {
if mapping == nil {
return modelName, false
}
mappedModelName := mapping[modelName]
if mappedModelName != "" {
return mappedModelName, true
}
return modelName, false
}