feat: use given usage if available in stream mode

统一文心计费模型名称

common/conv/any.go

@@ -0,0 +1,6 @@
+package conv
+
+func AsString(v any) string {
+	str, _ := v.(string)
+	return str
+}

common/model-ratio.go

@@ -75,7 +75,7 @@ var ModelRatio = map[string]float64{
 	"ERNIE-Bot":       0.8572,     // ￥0.012 / 1k tokens
 	"ERNIE-Bot-turbo": 0.5715,     // ￥0.008 / 1k tokens
 	"ERNIE-Bot-4":     0.12 * RMB, // ￥0.12 / 1k tokens
-	"ERNIE-Bot-8k":    0.024 * RMB,
+	"ERNIE-Bot-8K":    0.024 * RMB,
 	"Embedding-V1":    0.1429, // ￥0.002 / 1k tokens
 	"bge-large-zh":    0.002 * RMB,
 	"bge-large-en":    0.002 * RMB,

relay/channel/ali/main.go

@@ -48,7 +48,10 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 			MaxTokens:         request.MaxTokens,
 			Temperature:       request.Temperature,
 			TopP:              request.TopP,
+			TopK:              request.TopK,
+			ResultFormat:      "message",
 		},
+		Tools: request.Tools,
 	}
 }
 
@@ -117,19 +120,11 @@ func embeddingResponseAli2OpenAI(response *EmbeddingResponse) *openai.EmbeddingR
 }
 
 func responseAli2OpenAI(response *ChatResponse) *openai.TextResponse {
-	choice := openai.TextResponseChoice{
-		Index: 0,
-		Message: model.Message{
-			Role:    "assistant",
-			Content: response.Output.Text,
-		},
-		FinishReason: response.Output.FinishReason,
-	}
 	fullTextResponse := openai.TextResponse{
 		Id:      response.RequestId,
 		Object:  "chat.completion",
 		Created: helper.GetTimestamp(),
-		Choices: []openai.TextResponseChoice{choice},
+		Choices: response.Output.Choices,
 		Usage: model.Usage{
 			PromptTokens:     response.Usage.InputTokens,
 			CompletionTokens: response.Usage.OutputTokens,
@@ -140,10 +135,14 @@ func responseAli2OpenAI(response *ChatResponse) *openai.TextResponse {
 }
 
 func streamResponseAli2OpenAI(aliResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
+	if len(aliResponse.Output.Choices) == 0 {
+		return nil
+	}
+	aliChoice := aliResponse.Output.Choices[0]
 	var choice openai.ChatCompletionsStreamResponseChoice
-	choice.Delta.Content = aliResponse.Output.Text
-	if aliResponse.Output.FinishReason != "null" {
-		finishReason := aliResponse.Output.FinishReason
+	choice.Delta = aliChoice.Message
+	if aliChoice.FinishReason != "null" {
+		finishReason := aliChoice.FinishReason
 		choice.FinishReason = &finishReason
 	}
 	response := openai.ChatCompletionsStreamResponse{
@@ -204,6 +203,9 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 				usage.TotalTokens = aliResponse.Usage.InputTokens + aliResponse.Usage.OutputTokens
 			}
 			response := streamResponseAli2OpenAI(&aliResponse)
+			if response == nil {
+				return true
+			}
 			//response.Choices[0].Delta.Content = strings.TrimPrefix(response.Choices[0].Delta.Content, lastResponseText)
 			//lastResponseText = aliResponse.Output.Text
 			jsonResponse, err := json.Marshal(response)
@@ -226,6 +228,7 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 }
 
 func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	ctx := c.Request.Context()
 	var aliResponse ChatResponse
 	responseBody, err := io.ReadAll(resp.Body)
 	if err != nil {
@@ -235,6 +238,7 @@ func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
+	logger.Debugf(ctx, "response body: %s\n", responseBody)
 	err = json.Unmarshal(responseBody, &aliResponse)
 	if err != nil {
 		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil

relay/channel/ali/model.go

@@ -1,5 +1,10 @@
 package ali
 
+import (
+	"github.com/songquanpeng/one-api/relay/channel/openai"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
 type Message struct {
 	Content string `json:"content"`
 	Role    string `json:"role"`
@@ -18,12 +23,14 @@ type Parameters struct {
 	IncrementalOutput bool    `json:"incremental_output,omitempty"`
 	MaxTokens         int     `json:"max_tokens,omitempty"`
 	Temperature       float64 `json:"temperature,omitempty"`
+	ResultFormat      string  `json:"result_format,omitempty"`
 }
 
 type ChatRequest struct {
-	Model      string     `json:"model"`
-	Input      Input      `json:"input"`
-	Parameters Parameters `json:"parameters,omitempty"`
+	Model      string       `json:"model"`
+	Input      Input        `json:"input"`
+	Parameters Parameters   `json:"parameters,omitempty"`
+	Tools      []model.Tool `json:"tools,omitempty"`
 }
 
 type EmbeddingRequest struct {
@@ -62,8 +69,9 @@ type Usage struct {
 }
 
 type Output struct {
-	Text         string `json:"text"`
-	FinishReason string `json:"finish_reason"`
+	//Text         string                      `json:"text"`
+	//FinishReason string                      `json:"finish_reason"`
+	Choices []openai.TextResponseChoice `json:"choices"`
 }
 
 type ChatResponse struct {

relay/channel/anthropic/main.go

@@ -38,6 +38,7 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 		MaxTokens:   textRequest.MaxTokens,
 		Temperature: textRequest.Temperature,
 		TopP:        textRequest.TopP,
+		TopK:        textRequest.TopK,
 		Stream:      textRequest.Stream,
 	}
 	if claudeRequest.MaxTokens == 0 {

relay/channel/openai/adaptor.go

@@ -70,8 +70,10 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
 		var responseText string
-		err, responseText, _ = StreamHandler(c, resp, meta.Mode)
-		usage = ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
+		err, responseText, usage = StreamHandler(c, resp, meta.Mode)
+		if usage == nil {
+			usage = ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
+		}
 	} else {
 		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
 	}

relay/channel/openai/main.go

@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"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/relay/constant"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -53,7 +54,7 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 						continue // just ignore the error
 					}
 					for _, choice := range streamResponse.Choices {
-						responseText += choice.Delta.Content
+						responseText += conv.AsString(choice.Delta.Content)
 					}
 					if streamResponse.Usage != nil {
 						usage = streamResponse.Usage

relay/channel/openai/model.go

@@ -118,12 +118,9 @@ type ImageResponse struct {
 }
 
 type ChatCompletionsStreamResponseChoice struct {
-	Index int `json:"index"`
-	Delta struct {
-		Content string `json:"content"`
-		Role    string `json:"role,omitempty"`
-	} `json:"delta"`
-	FinishReason *string `json:"finish_reason,omitempty"`
+	Index        int           `json:"index"`
+	Delta        model.Message `json:"delta"`
+	FinishReason *string       `json:"finish_reason,omitempty"`
 }
 
 type ChatCompletionsStreamResponse struct {

relay/channel/tencent/main.go

@@ -10,6 +10,7 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/conv"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/channel/openai"
@@ -129,7 +130,7 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 			}
 			response := streamResponseTencent2OpenAI(&TencentResponse)
 			if len(response.Choices) != 0 {
-				responseText += response.Choices[0].Delta.Content
+				responseText += conv.AsString(response.Choices[0].Delta.Content)
 			}
 			jsonResponse, err := json.Marshal(response)
 			if err != nil {

relay/channel/xunfei/main.go

@@ -26,7 +26,11 @@ import (
 
 func requestOpenAI2Xunfei(request model.GeneralOpenAIRequest, xunfeiAppId string, domain string) *ChatRequest {
 	messages := make([]Message, 0, len(request.Messages))
+	var lastToolCalls []model.Tool
 	for _, message := range request.Messages {
+		if message.ToolCalls != nil {
+			lastToolCalls = message.ToolCalls
+		}
 		messages = append(messages, Message{
 			Role:    message.Role,
 			Content: message.StringContent(),
@@ -39,9 +43,33 @@ func requestOpenAI2Xunfei(request model.GeneralOpenAIRequest, xunfeiAppId string
 	xunfeiRequest.Parameter.Chat.TopK = request.N
 	xunfeiRequest.Parameter.Chat.MaxTokens = request.MaxTokens
 	xunfeiRequest.Payload.Message.Text = messages
+	if len(lastToolCalls) != 0 {
+		for _, toolCall := range lastToolCalls {
+			xunfeiRequest.Payload.Functions.Text = append(xunfeiRequest.Payload.Functions.Text, toolCall.Function)
+		}
+	}
+
 	return &xunfeiRequest
 }
 
+func getToolCalls(response *ChatResponse) []model.Tool {
+	var toolCalls []model.Tool
+	if len(response.Payload.Choices.Text) == 0 {
+		return toolCalls
+	}
+	item := response.Payload.Choices.Text[0]
+	if item.FunctionCall == nil {
+		return toolCalls
+	}
+	toolCall := model.Tool{
+		Id:       fmt.Sprintf("call_%s", helper.GetUUID()),
+		Type:     "function",
+		Function: *item.FunctionCall,
+	}
+	toolCalls = append(toolCalls, toolCall)
+	return toolCalls
+}
+
 func responseXunfei2OpenAI(response *ChatResponse) *openai.TextResponse {
 	if len(response.Payload.Choices.Text) == 0 {
 		response.Payload.Choices.Text = []ChatResponseTextItem{
@@ -53,8 +81,9 @@ func responseXunfei2OpenAI(response *ChatResponse) *openai.TextResponse {
 	choice := openai.TextResponseChoice{
 		Index: 0,
 		Message: model.Message{
-			Role:    "assistant",
-			Content: response.Payload.Choices.Text[0].Content,
+			Role:      "assistant",
+			Content:   response.Payload.Choices.Text[0].Content,
+			ToolCalls: getToolCalls(response),
 		},
 		FinishReason: constant.StopFinishReason,
 	}
@@ -78,6 +107,7 @@ func streamResponseXunfei2OpenAI(xunfeiResponse *ChatResponse) *openai.ChatCompl
 	}
 	var choice openai.ChatCompletionsStreamResponseChoice
 	choice.Delta.Content = xunfeiResponse.Payload.Choices.Text[0].Content
+	choice.Delta.ToolCalls = getToolCalls(xunfeiResponse)
 	if xunfeiResponse.Payload.Choices.Status == 2 {
 		choice.FinishReason = &constant.StopFinishReason
 	}

relay/channel/xunfei/model.go

@@ -26,13 +26,18 @@ type ChatRequest struct {
 		Message struct {
 			Text []Message `json:"text"`
 		} `json:"message"`
+		Functions struct {
+			Text []model.Function `json:"text,omitempty"`
+		} `json:"functions"`
 	} `json:"payload"`
 }
 
 type ChatResponseTextItem struct {
-	Content string `json:"content"`
-	Role    string `json:"role"`
-	Index   int    `json:"index"`
+	Content      string          `json:"content"`
+	Role         string          `json:"role"`
+	Index        int             `json:"index"`
+	ContentType  string          `json:"content_type"`
+	FunctionCall *model.Function `json:"function_call"`
 }
 
 type ChatResponse struct {

relay/model/general.go

@@ -5,25 +5,29 @@ type ResponseFormat struct {
 }
 
 type GeneralOpenAIRequest struct {
-	Model            string          `json:"model,omitempty"`
 	Messages         []Message       `json:"messages,omitempty"`
-	Prompt           any             `json:"prompt,omitempty"`
-	Stream           bool            `json:"stream,omitempty"`
-	MaxTokens        int             `json:"max_tokens,omitempty"`
-	Temperature      float64         `json:"temperature,omitempty"`
-	TopP             float64         `json:"top_p,omitempty"`
-	N                int             `json:"n,omitempty"`
-	Input            any             `json:"input,omitempty"`
-	Instruction      string          `json:"instruction,omitempty"`
-	Size             string          `json:"size,omitempty"`
-	Functions        any             `json:"functions,omitempty"`
+	Model            string          `json:"model,omitempty"`
 	FrequencyPenalty float64         `json:"frequency_penalty,omitempty"`
+	MaxTokens        int             `json:"max_tokens,omitempty"`
+	N                int             `json:"n,omitempty"`
 	PresencePenalty  float64         `json:"presence_penalty,omitempty"`
 	ResponseFormat   *ResponseFormat `json:"response_format,omitempty"`
 	Seed             float64         `json:"seed,omitempty"`
-	Tools            any             `json:"tools,omitempty"`
+	Stream           bool            `json:"stream,omitempty"`
+	Temperature      float64         `json:"temperature,omitempty"`
+	TopP             float64         `json:"top_p,omitempty"`
+	TopK             int             `json:"top_k,omitempty"`
+	Tools            []Tool          `json:"tools,omitempty"`
 	ToolChoice       any             `json:"tool_choice,omitempty"`
+	FunctionCall     any             `json:"function_call,omitempty"`
+	Functions        any             `json:"functions,omitempty"`
 	User             string          `json:"user,omitempty"`
+	Prompt           any             `json:"prompt,omitempty"`
+	Input            any             `json:"input,omitempty"`
+	EncodingFormat   string          `json:"encoding_format,omitempty"`
+	Dimensions       int             `json:"dimensions,omitempty"`
+	Instruction      string          `json:"instruction,omitempty"`
+	Size             string          `json:"size,omitempty"`
 }
 
 func (r GeneralOpenAIRequest) ParseInput() []string {

relay/model/message.go

@@ -1,9 +1,10 @@
 package model
 
 type Message struct {
-	Role    string  `json:"role"`
-	Content any     `json:"content"`
-	Name    *string `json:"name,omitempty"`
+	Role      string  `json:"role,omitempty"`
+	Content   any     `json:"content,omitempty"`
+	Name      *string `json:"name,omitempty"`
+	ToolCalls []Tool  `json:"tool_calls,omitempty"`
 }
 
 func (m Message) IsStringContent() bool {

relay/model/tool.go

@@ -0,0 +1,14 @@
+package model
+
+type Tool struct {
+	Id       string   `json:"id,omitempty"`
+	Type     string   `json:"type"`
+	Function Function `json:"function"`
+}
+
+type Function struct {
+	Description string `json:"description,omitempty"`
+	Name        string `json:"name"`
+	Parameters  any    `json:"parameters,omitempty"` // request
+	Arguments   any    `json:"arguments,omitempty"`  // response
+}