huangzhenpc/newapi-yx-diy
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: refactor token estimation logic

Browse Source 
- Introduced new OpenAI text models in `common/model.go`.
- Added `IsOpenAITextModel` function to check for OpenAI text models.
- Refactored token estimation methods across various channels to use estimated prompt tokens instead of direct prompt token counts.
- Updated related functions and structures to accommodate the new token estimation approach, enhancing overall token management.
This commit is contained in:
CaIon
2025-12-02 21:34:39 +08:00
parent d3ca454c3b
commit f5b409d74f
26 changed files with 389 additions and 275 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
relay/channel/claude/relay-claude.go
View File

@@ -673,7 +673,7 @@ func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claud
func HandleStreamFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, requestMode int) {
if requestMode == RequestModeCompletion {
claudeInfo.Usage = service.ResponseText2Usage(c, claudeInfo.ResponseText.String(), info.UpstreamModelName, info.PromptTokens)
claudeInfo.Usage = service.ResponseText2Usage(c, claudeInfo.ResponseText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
} else {
if claudeInfo.Usage.PromptTokens == 0 {
//上游出错
@@ -734,10 +734,7 @@ func HandleClaudeResponseData(c *gin.Context, info *relaycommon.RelayInfo, claud
return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
}
if requestMode == RequestModeCompletion {
completionTokens := service.CountTextToken(claudeResponse.Completion, info.OriginModelName)
claudeInfo.Usage.PromptTokens = info.PromptTokens
claudeInfo.Usage.CompletionTokens = completionTokens
claudeInfo.Usage.TotalTokens = info.PromptTokens + completionTokens
claudeInfo.Usage = service.ResponseText2Usage(c, claudeResponse.Completion, info.UpstreamModelName, info.GetEstimatePromptTokens())
} else {
claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens

10
relay/channel/cloudflare/relay_cloudflare.go
View File

@@ -74,7 +74,7 @@ func cfStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Res
if err := scanner.Err(); err != nil {
logger.LogError(c, "error_scanning_stream_response: "+err.Error())
}
usage := service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens)
usage := service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens())
if info.ShouldIncludeUsage {
response := helper.GenerateFinalUsageResponse(id, info.StartTime.Unix(), info.UpstreamModelName, *usage)
err := helper.ObjectData(c, response)
@@ -105,7 +105,7 @@ func cfHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response)
for _, choice := range response.Choices {
responseText += choice.Message.StringContent()
}
usage := service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens)
usage := service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens())
response.Usage = *usage
response.Id = helper.GetResponseID(c)
jsonResponse, err := json.Marshal(response)
@@ -142,10 +142,6 @@ func cfSTTHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Respon
c.Writer.WriteHeader(resp.StatusCode)
_, _ = c.Writer.Write(jsonResponse)
usage := &dto.Usage{}
usage.PromptTokens = info.PromptTokens
usage.CompletionTokens = service.CountTextToken(cfResp.Result.Text, info.UpstreamModelName)
usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
usage := service.ResponseText2Usage(c, cfResp.Result.Text, info.UpstreamModelName, info.GetEstimatePromptTokens())
return nil, usage
}

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

@@ -165,7 +165,7 @@ func cohereStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http
}
})
if usage.PromptTokens == 0 {
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens())
}
return usage, nil
}
@@ -225,9 +225,9 @@ func cohereRerankHandler(c *gin.Context, resp *http.Response, info *relaycommon.
}
usage := dto.Usage{}
if cohereResp.Meta.BilledUnits.InputTokens == 0 {
usage.PromptTokens = info.PromptTokens
usage.PromptTokens = info.GetEstimatePromptTokens()
usage.CompletionTokens = 0
usage.TotalTokens = info.PromptTokens
usage.TotalTokens = info.GetEstimatePromptTokens()
} else {
usage.PromptTokens = cohereResp.Meta.BilledUnits.InputTokens
usage.CompletionTokens = cohereResp.Meta.BilledUnits.OutputTokens

2
relay/channel/dify/relay-dify.go
View File

@@ -246,7 +246,7 @@ func difyStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.R
})
helper.Done(c)
if usage.TotalTokens == 0 {
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens())
}
usage.CompletionTokens += nodeToken
return usage, nil

8
relay/channel/gemini/relay-gemini-native.go
View File

@@ -5,7 +5,6 @@ import (
"net/http"
"github.com/QuantumNous/new-api/common"
"github.com/QuantumNous/new-api/constant"
"github.com/QuantumNous/new-api/dto"
"github.com/QuantumNous/new-api/logger"
relaycommon "github.com/QuantumNous/new-api/relay/common"
@@ -70,12 +69,7 @@ func NativeGeminiEmbeddingHandler(c *gin.Context, resp *http.Response, info *rel
println(string(responseBody))
}
usage := &dto.Usage{
PromptTokens: info.PromptTokens,
TotalTokens: info.PromptTokens,
}
common.SetContextKey(c, constant.ContextKeyLocalCountTokens, true)
usage := service.ResponseText2Usage(c, "", info.UpstreamModelName, info.GetEstimatePromptTokens())
if info.IsGeminiBatchEmbedding {
var geminiResponse dto.GeminiBatchEmbeddingResponse

8
relay/channel/gemini/relay-gemini.go
View File

@@ -1115,7 +1115,7 @@ func geminiStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http
if usage.CompletionTokens <= 0 {
str := responseText.String()
if len(str) > 0 {
usage = service.ResponseText2Usage(c, responseText.String(), info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
} else {
usage = &dto.Usage{}
}
@@ -1288,11 +1288,7 @@ func GeminiEmbeddingHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *h
// Google has not yet clarified how embedding models will be billed
// refer to openai billing method to use input tokens billing
// https://platform.openai.com/docs/guides/embeddings#what-are-embeddings
usage := &dto.Usage{
PromptTokens: info.PromptTokens,
CompletionTokens: 0,
TotalTokens: info.PromptTokens,
}
usage := service.ResponseText2Usage(c, "", info.UpstreamModelName, info.GetEstimatePromptTokens())
openAIResponse.Usage = *usage
jsonResponse, jsonErr := common.Marshal(openAIResponse)

2
relay/channel/minimax/tts.go
View File

@@ -163,7 +163,7 @@ func handleTTSResponse(c *gin.Context, resp *http.Response, info *relaycommon.Re
}
usage = &dto.Usage{
PromptTokens: info.PromptTokens,
PromptTokens: info.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: int(minimaxResp.ExtraInfo.UsageCharacters),
}

12
relay/channel/openai/relay-openai.go
View File

@@ -183,7 +183,7 @@ func OaiStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Re
}
if !containStreamUsage {
usage = service.ResponseText2Usage(c, responseTextBuilder.String(), info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseTextBuilder.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
usage.CompletionTokens += toolCount * 7
}
@@ -245,9 +245,9 @@ func OpenaiHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Respo
}
}
simpleResponse.Usage = dto.Usage{
PromptTokens: info.PromptTokens,
PromptTokens: info.GetEstimatePromptTokens(),
CompletionTokens: completionTokens,
TotalTokens: info.PromptTokens + completionTokens,
TotalTokens: info.GetEstimatePromptTokens() + completionTokens,
}
usageModified = true
}
@@ -336,8 +336,8 @@ func OpenaiTTSHandler(c *gin.Context, resp *http.Response, info *relaycommon.Rel
// and can be terminated directly.
defer service.CloseResponseBodyGracefully(resp)
usage := &dto.Usage{}
usage.PromptTokens = info.PromptTokens
usage.TotalTokens = info.PromptTokens
usage.PromptTokens = info.GetEstimatePromptTokens()
usage.TotalTokens = info.GetEstimatePromptTokens()
for k, v := range resp.Header {
c.Writer.Header().Set(k, v[0])
}
@@ -383,7 +383,7 @@ func OpenaiSTTHandler(c *gin.Context, resp *http.Response, info *relaycommon.Rel
}
usage := &dto.Usage{}
usage.PromptTokens = info.PromptTokens
usage.PromptTokens = info.GetEstimatePromptTokens()
usage.CompletionTokens = 0
usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
return nil, usage

2
relay/channel/openai/relay_responses.go
View File

@@ -141,7 +141,7 @@ func OaiResponsesStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp
}
if usage.PromptTokens == 0 && usage.CompletionTokens != 0 {
usage.PromptTokens = info.PromptTokens
usage.PromptTokens = info.GetEstimatePromptTokens()
}
usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens

2
relay/channel/palm/adaptor.go
View File

@@ -81,7 +81,7 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycom
if info.IsStream {
var responseText string
err, responseText = palmStreamHandler(c, resp)
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens())
} else {
usage, err = palmHandler(c, info, resp)
}

11
relay/channel/palm/relay-palm.go
View File

@@ -121,13 +121,8 @@ func palmHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Respons
}, resp.StatusCode)
}
fullTextResponse := responsePaLM2OpenAI(&palmResponse)
completionTokens := service.CountTextToken(palmResponse.Candidates[0].Content, info.UpstreamModelName)
usage := dto.Usage{
PromptTokens: info.PromptTokens,
CompletionTokens: completionTokens,
TotalTokens: info.PromptTokens + completionTokens,
}
fullTextResponse.Usage = usage
usage := service.ResponseText2Usage(c, palmResponse.Candidates[0].Content, info.UpstreamModelName, info.GetEstimatePromptTokens())
fullTextResponse.Usage = *usage
jsonResponse, err := common.Marshal(fullTextResponse)
if err != nil {
return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
@@ -135,5 +130,5 @@ func palmHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Respons
c.Writer.Header().Set("Content-Type", "application/json")
c.Writer.WriteHeader(resp.StatusCode)
service.IOCopyBytesGracefully(c, resp, jsonResponse)
return &usage, nil
return usage, nil
}

4
relay/channel/tencent/relay-tencent.go
View File

@@ -105,7 +105,7 @@ func tencentStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *htt
data = strings.TrimPrefix(data, "data:")
var tencentResponse TencentChatResponse
err := json.Unmarshal([]byte(data), &tencentResponse)
err := common.Unmarshal([]byte(data), &tencentResponse)
if err != nil {
common.SysLog("error unmarshalling stream response: " + err.Error())
continue
@@ -130,7 +130,7 @@ func tencentStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *htt
service.CloseResponseBodyGracefully(resp)
return service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.PromptTokens), nil
return service.ResponseText2Usage(c, responseText, info.UpstreamModelName, info.GetEstimatePromptTokens()), nil
}
func tencentHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {

12
relay/channel/volcengine/tts.go
View File

@@ -184,9 +184,9 @@ func handleTTSResponse(c *gin.Context, resp *http.Response, info *relaycommon.Re
c.Data(http.StatusOK, contentType, audioData)
usage = &dto.Usage{
PromptTokens: info.PromptTokens,
PromptTokens: info.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: info.PromptTokens,
TotalTokens: info.GetEstimatePromptTokens(),
}
return usage, nil
@@ -284,9 +284,9 @@ func handleTTSWebSocketResponse(c *gin.Context, requestURL string, volcRequest V
if msg.Sequence < 0 {
c.Status(http.StatusOK)
usage = &dto.Usage{
PromptTokens: info.PromptTokens,
PromptTokens: info.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: info.PromptTokens,
TotalTokens: info.GetEstimatePromptTokens(),
}
return usage, nil
}
@@ -297,9 +297,9 @@ func handleTTSWebSocketResponse(c *gin.Context, requestURL string, volcRequest V
c.Status(http.StatusOK)
usage = &dto.Usage{
PromptTokens: info.PromptTokens,
PromptTokens: info.GetEstimatePromptTokens(),
CompletionTokens: 0,
TotalTokens: info.PromptTokens,
TotalTokens: info.GetEstimatePromptTokens(),
}
return usage, nil
}

2
relay/channel/xai/text.go
View File

@@ -70,7 +70,7 @@ func xAIStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Re
})
if !containStreamUsage {
usage = service.ResponseText2Usage(c, responseTextBuilder.String(), info.UpstreamModelName, info.PromptTokens)
usage = service.ResponseText2Usage(c, responseTextBuilder.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
usage.CompletionTokens += toolCount * 7
}