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

Merge pull request #826 from Calcium-Ion/cache

Browse Source 
feat: Add prompt cache hit tokens support for DeepSeek channel #406
This commit is contained in:
Calcium-Ion
2025-03-08 16:52:19 +08:00
committed by GitHub
parent 618908f6f8 8c209e2fb9
commit d9390ff4c3
5 changed files with 26 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
dto/openai_response.go
View File

@@ -166,6 +166,7 @@ type Usage struct {
PromptTokens int `json:"prompt_tokens"` PromptTokens int `json:"prompt_tokens"`
CompletionTokens int `json:"completion_tokens"` CompletionTokens int `json:"completion_tokens"`
TotalTokens int `json:"total_tokens"` TotalTokens int `json:"total_tokens"`
PromptCacheHitTokens int `json:"prompt_cache_hit_tokens,omitempty"`
PromptTokensDetails InputTokenDetails `json:"prompt_tokens_details"` PromptTokensDetails InputTokenDetails `json:"prompt_tokens_details"`
CompletionTokenDetails OutputTokenDetails `json:"completion_tokens_details"` CompletionTokenDetails OutputTokenDetails `json:"completion_tokens_details"`
} }

6
relay/channel/openai/relay-openai.go
View File

@@ -254,6 +254,12 @@ func OaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.Rel
if !containStreamUsage { if !containStreamUsage {
usage, _ = service.ResponseText2Usage(responseTextBuilder.String(), info.UpstreamModelName, info.PromptTokens) usage, _ = service.ResponseText2Usage(responseTextBuilder.String(), info.UpstreamModelName, info.PromptTokens)
usage.CompletionTokens += toolCount * 7 usage.CompletionTokens += toolCount * 7
} else {
if info.ChannelType == common.ChannelTypeDeepSeek {
if usage.PromptCacheHitTokens != 0 {
usage.PromptTokensDetails.CachedTokens = usage.PromptCacheHitTokens
}
}
} }
if info.ShouldIncludeUsage && !containStreamUsage { if info.ShouldIncludeUsage && !containStreamUsage {

17
relay/relay-text.go
View File

@@ -320,19 +320,20 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
groupRatio := priceData.GroupRatio groupRatio := priceData.GroupRatio
modelPrice := priceData.ModelPrice modelPrice := priceData.ModelPrice
quota := 0 quotaCalculate := 0.0
if !priceData.UsePrice { if !priceData.UsePrice {
quota = (promptTokens - cacheTokens) + int(math.Round(float64(cacheTokens)*cacheRatio)) quotaCalculate = float64(promptTokens-cacheTokens) + float64(cacheTokens)*cacheRatio
quota += int(math.Round(float64(completionTokens) * completionRatio)) quotaCalculate += float64(completionTokens) * completionRatio
quota = int(math.Round(float64(quota) * ratio)) quotaCalculate = quotaCalculate * ratio
if ratio != 0 && quota <= 0 { if ratio != 0 && quotaCalculate <= 0 {
quota = 1 quotaCalculate = 1
} }
} else { } else {
quota = int(modelPrice * common.QuotaPerUnit * groupRatio) quotaCalculate = modelPrice * common.QuotaPerUnit * groupRatio
} }
quota := int(quotaCalculate)
totalTokens := promptTokens + completionTokens totalTokens := promptTokens + completionTokens
var logContent string var logContent string
if !priceData.UsePrice { if !priceData.UsePrice {
logContent = fmt.Sprintf("模型倍率 %.2f，补全倍率 %.2f，分组倍率 %.2f", modelRatio, completionRatio, groupRatio) logContent = fmt.Sprintf("模型倍率 %.2f，补全倍率 %.2f，分组倍率 %.2f", modelRatio, completionRatio, groupRatio)

13
service/quota.go
View File

@@ -4,7 +4,6 @@ import (
"errors" "errors"
"fmt" "fmt"
"github.com/bytedance/gopkg/util/gopool" "github.com/bytedance/gopkg/util/gopool"
"math"
"one-api/common" "one-api/common"
constant2 "one-api/constant" constant2 "one-api/constant"
"one-api/dto" "one-api/dto"
@@ -44,16 +43,18 @@ func calculateAudioQuota(info QuotaInfo) int {
audioCompletionRatio := operation_setting.GetAudioCompletionRatio(info.ModelName) audioCompletionRatio := operation_setting.GetAudioCompletionRatio(info.ModelName)
ratio := info.GroupRatio * info.ModelRatio ratio := info.GroupRatio * info.ModelRatio
quota := info.InputDetails.TextTokens + int(math.Round(float64(info.OutputDetails.TextTokens)*completionRatio)) quota := 0.0
quota += int(math.Round(float64(info.InputDetails.AudioTokens)*audioRatio)) + quota += float64(info.InputDetails.TextTokens)
int(math.Round(float64(info.OutputDetails.AudioTokens)*audioRatio*audioCompletionRatio)) quota += float64(info.OutputDetails.TextTokens) * completionRatio
quota += float64(info.InputDetails.AudioTokens) * audioRatio
quota += float64(info.OutputDetails.AudioTokens) * audioRatio * audioCompletionRatio
quota = int(math.Round(float64(quota) * ratio)) quota = quota * ratio
if ratio != 0 && quota <= 0 { if ratio != 0 && quota <= 0 {
quota = 1 quota = 1
} }
return quota return int(quota)
} }
func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.RealtimeUsage) error { func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.RealtimeUsage) error {

6
setting/operation_setting/cache_ratio.go
View File

@@ -16,9 +16,9 @@ var defaultCacheRatio = map[string]float64{
"gpt-4o-mini-2024-07-18": 0.5, "gpt-4o-mini-2024-07-18": 0.5,
"gpt-4o-realtime-preview": 0.5, "gpt-4o-realtime-preview": 0.5,
"gpt-4o-mini-realtime-preview": 0.5, "gpt-4o-mini-realtime-preview": 0.5,
"deepseek-chat": 0.5, "deepseek-chat": 0.1,
"deepseek-reasoner": 0.5, "deepseek-reasoner": 0.1,
"deepseek-coder": 0.5, "deepseek-coder": 0.1,
} }
var defaultCreateCacheRatio = map[string]float64{} var defaultCreateCacheRatio = map[string]float64{}