xinyin025/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-11-09 18:23:40 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

feat: support OpenRouter reasoning

Browse Source
This commit is contained in:
Laisky.Cai
2025-02-19 01:11:46 +00:00
parent 2a5908586d
commit a5f5e85c44
7 changed files with 80 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
relay/adaptor/openai/adaptor.go
View File

@@ -17,6 +17,7 @@ import (
"github.com/songquanpeng/one-api/relay/adaptor/geminiv2"
"github.com/songquanpeng/one-api/relay/adaptor/minimax"
"github.com/songquanpeng/one-api/relay/adaptor/novita"
"github.com/songquanpeng/one-api/relay/adaptor/openrouter"
"github.com/songquanpeng/one-api/relay/channeltype"
"github.com/songquanpeng/one-api/relay/meta"
"github.com/songquanpeng/one-api/relay/model"
@@ -95,6 +96,21 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
return nil, errors.New("request is nil")
}
meta := meta.GetByContext(c)
switch meta.ChannelType {
case channeltype.OpenRouter:
includeReasoning := true
request.IncludeReasoning = &includeReasoning
if request.Provider == nil || request.Provider.Sort == "" {
if request.Provider == nil {
request.Provider = &openrouter.RequestProvider{}
}
request.Provider.Sort = "throughput"
}
default:
}
if request.Stream && !config.EnforceIncludeUsage {
logger.Warn(c.Request.Context(),
"please set ENFORCE_INCLUDE_USAGE=true to ensure accurate billing in stream mode")

35
relay/adaptor/openai/main.go
View File

@@ -27,6 +27,7 @@ const (
func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
responseText := ""
reasoningText := ""
scanner := bufio.NewScanner(resp.Body)
scanner.Split(bufio.ScanLines)
var usage *model.Usage
@@ -62,6 +63,10 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
}
render.StringData(c, data)
for _, choice := range streamResponse.Choices {
if choice.Delta.Reasoning != nil {
reasoningText += *choice.Delta.Reasoning
}
responseText += conv.AsString(choice.Delta.Content)
}
if streamResponse.Usage != nil {
@@ -94,7 +99,7 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", nil
}
return nil, responseText, usage
return nil, reasoningText + responseText, usage
}
// Handler handles the non-stream response from OpenAI API
@@ -150,20 +155,26 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
CompletionTokens: completionTokens,
TotalTokens: promptTokens + completionTokens,
}
} else if textResponse.PromptTokensDetails.AudioTokens+textResponse.CompletionTokensDetails.AudioTokens > 0 {
} else if (textResponse.PromptTokensDetails != nil && textResponse.PromptTokensDetails.AudioTokens > 0) ||
(textResponse.CompletionTokensDetails != nil && textResponse.CompletionTokensDetails.AudioTokens > 0) {
// Convert the more expensive audio tokens to uniformly priced text tokens.
// Note that when there are no audio tokens in prompt and completion,
// OpenAI will return empty PromptTokensDetails and CompletionTokensDetails, which can be misleading.
textResponse.Usage.PromptTokens = textResponse.PromptTokensDetails.TextTokens +
int(math.Ceil(
float64(textResponse.PromptTokensDetails.AudioTokens)*
ratio.GetAudioPromptRatio(modelName),
))
textResponse.Usage.CompletionTokens = textResponse.CompletionTokensDetails.TextTokens +
int(math.Ceil(
float64(textResponse.CompletionTokensDetails.AudioTokens)*
ratio.GetAudioPromptRatio(modelName)*ratio.GetAudioCompletionRatio(modelName),
))
if textResponse.PromptTokensDetails != nil {
textResponse.Usage.PromptTokens = textResponse.PromptTokensDetails.TextTokens +
int(math.Ceil(
float64(textResponse.PromptTokensDetails.AudioTokens)*
ratio.GetAudioPromptRatio(modelName),
))
}
if textResponse.CompletionTokensDetails != nil {
textResponse.Usage.CompletionTokens = textResponse.CompletionTokensDetails.TextTokens +
int(math.Ceil(
float64(textResponse.CompletionTokensDetails.AudioTokens)*
ratio.GetAudioPromptRatio(modelName)*ratio.GetAudioCompletionRatio(modelName),
))
}
textResponse.Usage.TotalTokens = textResponse.Usage.PromptTokens +
textResponse.Usage.CompletionTokens