mirror of
https://github.com/songquanpeng/one-api.git
synced 2025-11-08 01:33:43 +08:00
Merge remote-tracking branch 'origin/upstream/main'
This commit is contained in:
@@ -3,8 +3,12 @@ package controller
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"github.com/Laisky/errors/v2"
|
||||
"math"
|
||||
"net/http"
|
||||
"strings"
|
||||
|
||||
"github.com/gin-gonic/gin"
|
||||
"github.com/pkg/errors"
|
||||
"github.com/songquanpeng/one-api/common"
|
||||
"github.com/songquanpeng/one-api/common/config"
|
||||
"github.com/songquanpeng/one-api/common/logger"
|
||||
@@ -12,13 +16,11 @@ import (
|
||||
"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/constant/role"
|
||||
"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/relaymode"
|
||||
"math"
|
||||
"net/http"
|
||||
"strings"
|
||||
)
|
||||
|
||||
func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.GeneralOpenAIRequest, error) {
|
||||
@@ -89,7 +91,7 @@ func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIR
|
||||
return preConsumedQuota, nil
|
||||
}
|
||||
|
||||
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64) (quota int64) {
|
||||
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64, systemPromptReset bool) {
|
||||
if usage == nil {
|
||||
logger.Error(ctx, "usage is nil, which is unexpected")
|
||||
return
|
||||
@@ -98,7 +100,7 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
|
||||
completionRatio := billingratio.GetCompletionRatio(textRequest.Model, meta.ChannelType)
|
||||
promptTokens := usage.PromptTokens
|
||||
completionTokens := usage.CompletionTokens
|
||||
quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
|
||||
quota := int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
|
||||
if ratio != 0 && quota <= 0 {
|
||||
quota = 1
|
||||
}
|
||||
@@ -117,8 +119,11 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
|
||||
if err != nil {
|
||||
logger.Error(ctx, "error update user quota cache: "+err.Error())
|
||||
}
|
||||
|
||||
logContent := fmt.Sprintf("模型倍率 %.2f,分组倍率 %.2f,补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
|
||||
var extraLog string
|
||||
if systemPromptReset {
|
||||
extraLog = " (注意系统提示词已被重置)"
|
||||
}
|
||||
logContent := fmt.Sprintf("模型倍率 %.2f,分组倍率 %.2f,补全倍率 %.2f%s", modelRatio, groupRatio, completionRatio, extraLog)
|
||||
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)
|
||||
@@ -156,3 +161,23 @@ func isErrorHappened(meta *meta.Meta, resp *http.Response) bool {
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func setSystemPrompt(ctx context.Context, request *relaymodel.GeneralOpenAIRequest, prompt string) (reset bool) {
|
||||
if prompt == "" {
|
||||
return false
|
||||
}
|
||||
if len(request.Messages) == 0 {
|
||||
return false
|
||||
}
|
||||
if request.Messages[0].Role == role.System {
|
||||
request.Messages[0].Content = prompt
|
||||
logger.Infof(ctx, "rewrite system prompt")
|
||||
return true
|
||||
}
|
||||
request.Messages = append([]relaymodel.Message{{
|
||||
Role: role.System,
|
||||
Content: prompt,
|
||||
}}, request.Messages...)
|
||||
logger.Infof(ctx, "add system prompt")
|
||||
return true
|
||||
}
|
||||
|
||||
@@ -8,9 +8,9 @@ import (
|
||||
|
||||
"github.com/Laisky/errors/v2"
|
||||
"github.com/gin-gonic/gin"
|
||||
"github.com/songquanpeng/one-api/common/config"
|
||||
"github.com/songquanpeng/one-api/common/ctxkey"
|
||||
"github.com/songquanpeng/one-api/common/logger"
|
||||
"github.com/songquanpeng/one-api/model"
|
||||
"github.com/songquanpeng/one-api/relay"
|
||||
"github.com/songquanpeng/one-api/relay/adaptor"
|
||||
"github.com/songquanpeng/one-api/relay/adaptor/openai"
|
||||
@@ -37,6 +37,8 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
|
||||
meta.OriginModelName = textRequest.Model
|
||||
textRequest.Model, _ = getMappedModelName(textRequest.Model, meta.ModelMapping)
|
||||
meta.ActualModelName = textRequest.Model
|
||||
// set system prompt if not empty
|
||||
systemPromptReset := setSystemPrompt(ctx, textRequest, meta.SystemPrompt)
|
||||
// get model ratio & group ratio
|
||||
modelRatio := billingratio.GetModelRatio(textRequest.Model, meta.ChannelType)
|
||||
// groupRatio := billingratio.GetGroupRatio(meta.Group)
|
||||
@@ -88,23 +90,12 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
|
||||
}
|
||||
|
||||
// post-consume quota
|
||||
go func() {
|
||||
quota := postConsumeQuota(c, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio)
|
||||
docu := model.NewUserRequestCost(
|
||||
c.GetInt(ctxkey.Id),
|
||||
c.GetString(ctxkey.RequestId),
|
||||
quota,
|
||||
)
|
||||
if err = docu.Insert(); err != nil {
|
||||
logger.Errorf(c, "insert user request cost failed: %+v", err)
|
||||
}
|
||||
}()
|
||||
|
||||
go postConsumeQuota(ctx, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio, systemPromptReset)
|
||||
return nil
|
||||
}
|
||||
|
||||
func getRequestBody(c *gin.Context, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, adaptor adaptor.Adaptor) (io.Reader, error) {
|
||||
if meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
|
||||
if !config.EnforceIncludeUsage && meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
|
||||
// no need to convert request for openai
|
||||
return c.Request.Body, nil
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user