Merge 7ec33793b7 into 7ac553541b

common/config/config.go

@@ -164,3 +164,6 @@ var UserContentRequestTimeout = env.Int("USER_CONTENT_REQUEST_TIMEOUT", 30)
 
 var EnforceIncludeUsage = env.Bool("ENFORCE_INCLUDE_USAGE", false)
 var TestPrompt = env.String("TEST_PROMPT", "Output only your specific model name with no additional text.")
+
+// OpenrouterProviderSort is used to determine the order of the providers in the openrouter
+var OpenrouterProviderSort = env.String("OPENROUTER_PROVIDER_SORT", "")

common/conv/any.go

@@ -1,6 +1,9 @@
 package conv
 
 func AsString(v any) string {
-	str, _ := v.(string)
-	return str
+	if str, ok := v.(string); ok {
+		return str
+	}
+
+	return ""
 }

relay/adaptor/openai/adaptor.go

@@ -9,6 +9,8 @@ import (
 
 	"github.com/gin-gonic/gin"
 
+	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/alibailian"
 	"github.com/songquanpeng/one-api/relay/adaptor/baiduv2"
@@ -16,6 +18,7 @@ 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"
@@ -85,7 +88,29 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
-	if request.Stream {
+
+	meta := meta.GetByContext(c)
+	switch meta.ChannelType {
+	case channeltype.OpenRouter:
+		includeReasoning := true
+		request.IncludeReasoning = &includeReasoning
+		if request.Provider == nil || request.Provider.Sort == "" &&
+			config.OpenrouterProviderSort != "" {
+			if request.Provider == nil {
+				request.Provider = &openrouter.RequestProvider{}
+			}
+
+			request.Provider.Sort = config.OpenrouterProviderSort
+		}
+	default:
+	}
+
+	if request.Stream && !config.EnforceIncludeUsage {
+		logger.Warn(c.Request.Context(),
+			"please set ENFORCE_INCLUDE_USAGE=true to ensure accurate billing in stream mode")
+	}
+
+	if config.EnforceIncludeUsage && request.Stream {
 		// always return usage in stream mode
 		if request.StreamOptions == nil {
 			request.StreamOptions = &model.StreamOptions{}

relay/adaptor/openai/main.go

@@ -8,12 +8,11 @@ import (
 	"net/http"
 	"strings"
 
-	"github.com/songquanpeng/one-api/common/render"
-
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/conv"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/common/render"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 )
@@ -26,6 +25,7 @@ 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
@@ -61,6 +61,13 @@ 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
+				}
+				if choice.Delta.ReasoningContent != nil {
+					reasoningText += *choice.Delta.ReasoningContent
+				}
+
 				responseText += conv.AsString(choice.Delta.Content)
 			}
 			if streamResponse.Usage != nil {
@@ -93,7 +100,7 @@ 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
 }
 
 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
@@ -136,10 +143,17 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
 
-	if textResponse.Usage.TotalTokens == 0 || (textResponse.Usage.PromptTokens == 0 && textResponse.Usage.CompletionTokens == 0) {
+	if textResponse.Usage.TotalTokens == 0 ||
+		(textResponse.Usage.PromptTokens == 0 && textResponse.Usage.CompletionTokens == 0) {
 		completionTokens := 0
 		for _, choice := range textResponse.Choices {
 			completionTokens += CountTokenText(choice.Message.StringContent(), modelName)
+			if choice.Message.Reasoning != nil {
+				completionTokens += CountToken(*choice.Message.Reasoning)
+			}
+			if choice.ReasoningContent != nil {
+				completionTokens += CountToken(*choice.ReasoningContent)
+			}
 		}
 		textResponse.Usage = model.Usage{
 			PromptTokens:     promptTokens,
@@ -147,5 +161,6 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 			TotalTokens:      promptTokens + completionTokens,
 		}
 	}
+
 	return nil, &textResponse.Usage
 }

relay/adaptor/openrouter/model.go

@@ -0,0 +1,22 @@
+package openrouter
+
+// RequestProvider customize how your requests are routed using the provider object
+// in the request body for Chat Completions and Completions.
+//
+// https://openrouter.ai/docs/features/provider-routing
+type RequestProvider struct {
+	// Order is list of provider names to try in order (e.g. ["Anthropic", "OpenAI"]). Default: empty
+	Order []string `json:"order,omitempty"`
+	// AllowFallbacks is whether to allow backup providers when the primary is unavailable. Default: true
+	AllowFallbacks bool `json:"allow_fallbacks,omitempty"`
+	// RequireParameters is only use providers that support all parameters in your request. Default: false
+	RequireParameters bool `json:"require_parameters,omitempty"`
+	// DataCollection is control whether to use providers that may store data ("allow" or "deny"). Default: "allow"
+	DataCollection string `json:"data_collection,omitempty" binding:"omitempty,oneof=allow deny"`
+	// Ignore is list of provider names to skip for this request. Default: empty
+	Ignore []string `json:"ignore,omitempty"`
+	// Quantizations is list of quantization levels to filter by (e.g. ["int4", "int8"]). Default: empty
+	Quantizations []string `json:"quantizations,omitempty"`
+	// Sort is sort providers by price or throughput (e.g. "price" or "throughput"). Default: empty
+	Sort string `json:"sort,omitempty" binding:"omitempty,oneof=price throughput latency"`
+}

relay/controller/helper.go

@@ -102,6 +102,9 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
 	var quota int64
 	completionRatio := billingratio.GetCompletionRatio(textRequest.Model, meta.ChannelType)
 	promptTokens := usage.PromptTokens
+	// It appears that DeepSeek's official service automatically merges ReasoningTokens into CompletionTokens,
+	// but the behavior of third-party providers may differ, so for now we do not add them manually.
+	// completionTokens := usage.CompletionTokens + usage.CompletionTokensDetails.ReasoningTokens
 	completionTokens := usage.CompletionTokens
 	quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
 	if ratio != 0 && quota <= 0 {

relay/model/general.go

@@ -1,5 +1,7 @@
 package model
 
+import "github.com/songquanpeng/one-api/relay/adaptor/openrouter"
+
 type ResponseFormat struct {
 	Type       string      `json:"type,omitempty"`
 	JsonSchema *JSONSchema `json:"json_schema,omitempty"`
@@ -66,6 +68,11 @@ type GeneralOpenAIRequest struct {
 	// Others
 	Instruction string `json:"instruction,omitempty"`
 	NumCtx      int    `json:"num_ctx,omitempty"`
+	// -------------------------------------
+	// Openrouter
+	// -------------------------------------
+	Provider         *openrouter.RequestProvider `json:"provider,omitempty"`
+	IncludeReasoning *bool                       `json:"include_reasoning,omitempty"`
 }
 
 func (r GeneralOpenAIRequest) ParseInput() []string {

relay/model/message.go

@@ -1,12 +1,35 @@
 package model
 
 type Message struct {
-	Role             string  `json:"role,omitempty"`
-	Content          any     `json:"content,omitempty"`
-	ReasoningContent any     `json:"reasoning_content,omitempty"`
-	Name             *string `json:"name,omitempty"`
-	ToolCalls        []Tool  `json:"tool_calls,omitempty"`
-	ToolCallId       string  `json:"tool_call_id,omitempty"`
+	Role string `json:"role,omitempty"`
+	// Content is a string or a list of objects
+	Content    any           `json:"content,omitempty"`
+	Name       *string       `json:"name,omitempty"`
+	ToolCalls  []Tool        `json:"tool_calls,omitempty"`
+	ToolCallId string        `json:"tool_call_id,omitempty"`
+	Audio      *messageAudio `json:"audio,omitempty"`
+	// -------------------------------------
+	// Deepseek 专有的一些字段
+	// https://api-docs.deepseek.com/api/create-chat-completion
+	// -------------------------------------
+	// Prefix forces the model to begin its answer with the supplied prefix in the assistant message.
+	// To enable this feature, set base_url to "https://api.deepseek.com/beta".
+	Prefix *bool `json:"prefix,omitempty"` // ReasoningContent is Used for the deepseek-reasoner model in the Chat
+	// Prefix Completion feature as the input for the CoT in the last assistant message.
+	// When using this feature, the prefix parameter must be set to true.
+	ReasoningContent *string `json:"reasoning_content,omitempty"`
+	// -------------------------------------
+	// Openrouter
+	// -------------------------------------
+	Reasoning *string `json:"reasoning,omitempty"`
+	Refusal   *bool   `json:"refusal,omitempty"`
+}
+
+type messageAudio struct {
+	Id         string `json:"id"`
+	Data       string `json:"data,omitempty"`
+	ExpiredAt  int    `json:"expired_at,omitempty"`
+	Transcript string `json:"transcript,omitempty"`
 }
 
 func (m Message) IsStringContent() bool {

relay/model/misc.go

@@ -4,14 +4,12 @@ type Usage struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
 	TotalTokens      int `json:"total_tokens"`
-
-	CompletionTokensDetails *CompletionTokensDetails `json:"completion_tokens_details,omitempty"`
-}
-
-type CompletionTokensDetails struct {
-	ReasoningTokens          int `json:"reasoning_tokens"`
-	AcceptedPredictionTokens int `json:"accepted_prediction_tokens"`
-	RejectedPredictionTokens int `json:"rejected_prediction_tokens"`
+	// PromptTokensDetails may be empty for some models
+	PromptTokensDetails *usagePromptTokensDetails `gorm:"-" json:"prompt_tokens_details,omitempty"`
+	// CompletionTokensDetails may be empty for some models
+	CompletionTokensDetails *usageCompletionTokensDetails `gorm:"-" json:"completion_tokens_details,omitempty"`
+	ServiceTier             string                        `gorm:"-" json:"service_tier,omitempty"`
+	SystemFingerprint       string                        `gorm:"-" json:"system_fingerprint,omitempty"`
 }
 
 type Error struct {
@@ -25,3 +23,20 @@ type ErrorWithStatusCode struct {
 	Error
 	StatusCode int `json:"status_code"`
 }
+
+type usagePromptTokensDetails struct {
+	CachedTokens int `json:"cached_tokens"`
+	AudioTokens  int `json:"audio_tokens"`
+	// TextTokens could be zero for pure text chats
+	TextTokens  int `json:"text_tokens"`
+	ImageTokens int `json:"image_tokens"`
+}
+
+type usageCompletionTokensDetails struct {
+	ReasoningTokens          int `json:"reasoning_tokens"`
+	AudioTokens              int `json:"audio_tokens"`
+	AcceptedPredictionTokens int `json:"accepted_prediction_tokens"`
+	RejectedPredictionTokens int `json:"rejected_prediction_tokens"`
+	// TextTokens could be zero for pure text chats
+	TextTokens int `json:"text_tokens"`
+}