Merge pull request #1122 from akkuman/feat/stream-tts

feat: streaming response for tts

README.en.md

@@ -110,6 +110,7 @@ For detailed configuration instructions, please refer to [Installation Guide-Env
 - `AZURE_DEFAULT_API_VERSION`: Azure channel default API version, default is `2025-04-01-preview`
 - `NOTIFICATION_LIMIT_DURATION_MINUTE`: Notification limit duration, default is `10` minutes
 - `NOTIFY_LIMIT_COUNT`: Maximum number of user notifications within the specified duration, default is `2`
+- `ERROR_LOG_ENABLED=true`: Whether to record and display error logs, default is `false`
 
 ## Deployment
 

README.md

@@ -110,6 +110,7 @@ New API提供了丰富的功能，详细特性请参考[特性说明](https://do
 - `AZURE_DEFAULT_API_VERSION`：Azure渠道默认API版本，默认 `2025-04-01-preview`
 - `NOTIFICATION_LIMIT_DURATION_MINUTE`：通知限制持续时间，默认 `10`分钟
 - `NOTIFY_LIMIT_COUNT`：用户通知在指定持续时间内的最大数量，默认 `2`
+- `ERROR_LOG_ENABLED=true`: 是否记录并显示错误日志，默认`false`
 
 ## 部署
 

controller/channel.go

@@ -119,8 +119,11 @@ func FetchUpstreamModels(c *gin.Context) {
 		baseURL = channel.GetBaseURL()
 	}
 	url := fmt.Sprintf("%s/v1/models", baseURL)
-	if channel.Type == common.ChannelTypeGemini {
+	switch channel.Type {
+	case common.ChannelTypeGemini:
 		url = fmt.Sprintf("%s/v1beta/openai/models", baseURL)
+	case common.ChannelTypeAli:
+		url = fmt.Sprintf("%s/compatible-mode/v1/models", baseURL)
 	}
 	body, err := GetResponseBody("GET", url, channel, GetAuthHeader(channel.Key))
 	if err != nil {

dto/openai_request.go

@@ -43,6 +43,7 @@ type GeneralOpenAIRequest struct {
 	ResponseFormat   *ResponseFormat   `json:"response_format,omitempty"`
 	EncodingFormat   any               `json:"encoding_format,omitempty"`
 	Seed             float64           `json:"seed,omitempty"`
+	ParallelTooCalls *bool             `json:"parallel_tool_calls,omitempty"`
 	Tools            []ToolCallRequest `json:"tools,omitempty"`
 	ToolChoice       any               `json:"tool_choice,omitempty"`
 	User             string            `json:"user,omitempty"`
@@ -53,6 +54,7 @@ type GeneralOpenAIRequest struct {
 	Audio            any               `json:"audio,omitempty"`
 	EnableThinking   any               `json:"enable_thinking,omitempty"` // ali
 	ExtraBody        any               `json:"extra_body,omitempty"`
+	WebSearchOptions *WebSearchOptions `json:"web_search_options,omitempty"`
 }
 
 type ToolCallRequest struct {
@@ -371,6 +373,11 @@ func (m *Message) ParseContent() []MediaContent {
 	return contentList
 }
 
+type WebSearchOptions struct {
+	SearchContextSize string          `json:"search_context_size,omitempty"`
+	UserLocation      json.RawMessage `json:"user_location,omitempty"`
+}
+
 type OpenAIResponsesRequest struct {
 	Model              string               `json:"model"`
 	Input              json.RawMessage      `json:"input,omitempty"`

relay/channel/ali/adaptor.go

@@ -57,6 +57,12 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
+
+	// fix: ali parameter.enable_thinking must be set to false for non-streaming calls
+	if !info.IsStream {
+		request.EnableThinking = false
+	}
+
 	switch info.RelayMode {
 	default:
 		aliReq := requestOpenAI2Ali(*request)

relay/channel/api_request.go

@@ -122,11 +122,13 @@ func doRequest(c *gin.Context, req *http.Request, info *common.RelayInfo) (*http
 	var pingerWg sync.WaitGroup
 	if info.IsStream {
 		helper.SetEventStreamHeaders(c)
-		pingInterval := time.Duration(generalSettings.PingIntervalSeconds) * time.Second
-		var pingerCtx context.Context
-		pingerCtx, stopPinger = context.WithCancel(c.Request.Context())
 
 		if pingEnabled {
+			pingInterval := time.Duration(generalSettings.PingIntervalSeconds) * time.Second
+			var pingerCtx context.Context
+			pingerCtx, stopPinger = context.WithCancel(c.Request.Context())
+			// 退出时清理 pingerCtx 防止泄露
+			defer stopPinger()
 			pingerWg.Add(1)
 			gopool.Go(func() {
 				defer pingerWg.Done()
@@ -166,9 +168,8 @@ func doRequest(c *gin.Context, req *http.Request, info *common.RelayInfo) (*http
 	}
 
 	resp, err := client.Do(req)
-	// request结束后停止ping
+	// request结束后等待 ping goroutine 完成
 	if info.IsStream && pingEnabled {
-		stopPinger()
 		pingerWg.Wait()
 	}
 	if err != nil {

relay/channel/gemini/relay-gemini.go

@@ -39,15 +39,22 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 	}
 
 	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
-		if strings.HasSuffix(info.OriginModelName, "-thinking") {
-			budgetTokens := model_setting.GetGeminiSettings().ThinkingAdapterBudgetTokensPercentage * float64(geminiRequest.GenerationConfig.MaxOutputTokens)
-			if budgetTokens == 0 || budgetTokens > 24576 {
-				budgetTokens = 24576
-			}
-			geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
-				ThinkingBudget:  common.GetPointer(int(budgetTokens)),
-				IncludeThoughts: true,
-			}
+	        if strings.HasSuffix(info.OriginModelName, "-thinking") {
+	            // 如果模型名以 gemini-2.5-pro 开头，不设置 ThinkingBudget
+	            if strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro") {
+	                geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+	                    IncludeThoughts: true,
+	                }
+	            } else {
+	                budgetTokens := model_setting.GetGeminiSettings().ThinkingAdapterBudgetTokensPercentage * float64(geminiRequest.GenerationConfig.MaxOutputTokens)
+	                if budgetTokens == 0 || budgetTokens > 24576 {
+	                    budgetTokens = 24576
+	                }
+	                geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+	                    ThinkingBudget:  common.GetPointer(int(budgetTokens)),
+	                    IncludeThoughts: true,
+	                }
+	            }
 		} else if strings.HasSuffix(info.OriginModelName, "-nothinking") {
 			geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
 				ThinkingBudget: common.GetPointer(0),

relay/channel/openai/relay-openai.go

@@ -273,36 +273,25 @@ func OpenaiHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayI
 }
 
 func OpenaiTTSHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {
-	responseBody, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
-	}
-	err = resp.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
-	}
-	// Reset response body
-	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
-	// We shouldn't set the header before we parse the response body, because the parse part may fail.
-	// And then we will have to send an error response, but in this case, the header has already been set.
-	// So the httpClient will be confused by the response.
-	// For example, Postman will report error, and we cannot check the response at all.
+	// the status code has been judged before, if there is a body reading failure,
+	// it should be regarded as a non-recoverable error, so it should not return err for external retry.
+	// Analogous to nginx's load balancing, it will only retry if it can't be requested or 
+	// if the upstream returns a specific status code, once the upstream has already written the header, 
+	// the subsequent failure of the response body should be regarded as a non-recoverable error, 
+	// and can be terminated directly.
+	defer resp.Body.Close()
+	usage := &dto.Usage{}
+	usage.PromptTokens = info.PromptTokens
+	usage.TotalTokens = info.PromptTokens
 	for k, v := range resp.Header {
 		c.Writer.Header().Set(k, v[0])
 	}
 	c.Writer.WriteHeader(resp.StatusCode)
-	_, err = io.Copy(c.Writer, resp.Body)
+	c.Writer.WriteHeaderNow()
+	_, err := io.Copy(c.Writer, resp.Body)
 	if err != nil {
-		return service.OpenAIErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError), nil
+		common.LogError(c, err.Error())
 	}
-	err = resp.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
-	}
-
-	usage := &dto.Usage{}
-	usage.PromptTokens = info.PromptTokens
-	usage.TotalTokens = info.PromptTokens
 	return nil, usage
 }
 

relay/channel/vertex/adaptor.go

@@ -95,14 +95,23 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 		} else {
 			suffix = "generateContent"
 		}
-		return fmt.Sprintf(
-			"https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/google/models/%s:%s",
-			region,
-			adc.ProjectID,
-			region,
-			info.UpstreamModelName,
-			suffix,
-		), nil
+		if region == "global" {
+			return fmt.Sprintf(
+				"https://aiplatform.googleapis.com/v1/projects/%s/locations/global/publishers/google/models/%s:%s",
+				adc.ProjectID,
+				info.UpstreamModelName,
+				suffix,
+			), nil
+		} else {
+			return fmt.Sprintf(
+				"https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/google/models/%s:%s",
+				region,
+				adc.ProjectID,
+				region,
+				info.UpstreamModelName,
+				suffix,
+			), nil
+		}
 	} else if a.RequestMode == RequestModeClaude {
 		if info.IsStream {
 			suffix = "streamRawPredict?alt=sse"

relay/relay-image.go

@@ -46,11 +46,23 @@ func getAndValidImageRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.
 		if err != nil {
 			return nil, err
 		}
+
+		if imageRequest.Model == "" {
+			imageRequest.Model = "dall-e-3"
+		}
+
+		if strings.Contains(imageRequest.Size, "×") {
+			return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
+		}
+
 		// Not "256x256", "512x512", or "1024x1024"
 		if imageRequest.Model == "dall-e-2" || imageRequest.Model == "dall-e" {
 			if imageRequest.Size != "" && imageRequest.Size != "256x256" && imageRequest.Size != "512x512" && imageRequest.Size != "1024x1024" {
 				return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024 for dall-e-2 or dall-e")
 			}
+			if imageRequest.Size == "" {
+				imageRequest.Size = "1024x1024"
+			}
 		} else if imageRequest.Model == "dall-e-3" {
 			if imageRequest.Size != "" && imageRequest.Size != "1024x1024" && imageRequest.Size != "1024x1792" && imageRequest.Size != "1792x1024" {
 				return nil, errors.New("size must be one of 1024x1024, 1024x1792 or 1792x1024 for dall-e-3")
@@ -58,74 +70,24 @@ func getAndValidImageRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.
 			if imageRequest.Quality == "" {
 				imageRequest.Quality = "standard"
 			}
-			// N should between 1 and 10
-			//if imageRequest.N != 0 && (imageRequest.N < 1 || imageRequest.N > 10) {
-			//	return service.OpenAIErrorWrapper(errors.New("n must be between 1 and 10"), "invalid_field_value", http.StatusBadRequest)
-			//}
+			if imageRequest.Size == "" {
+				imageRequest.Size = "1024x1024"
+			}
+		} else if imageRequest.Model == "gpt-image-1" {
+			if imageRequest.Quality == "" {
+				imageRequest.Quality = "auto"
+			}
+		}
+
+		if imageRequest.Prompt == "" {
+			return nil, errors.New("prompt is required")
+		}
+
+		if imageRequest.N == 0 {
+			imageRequest.N = 1
 		}
 	}
 
-	if imageRequest.Prompt == "" {
-		return nil, errors.New("prompt is required")
-	}
-
-	if imageRequest.Model == "" {
-		imageRequest.Model = "dall-e-2"
-	}
-	if strings.Contains(imageRequest.Size, "×") {
-		return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
-	}
-	if imageRequest.N == 0 {
-		imageRequest.N = 1
-	}
-	if imageRequest.Size == "" {
-		imageRequest.Size = "1024x1024"
-	}
-
-	err := common.UnmarshalBodyReusable(c, imageRequest)
-	if err != nil {
-		return nil, err
-	}
-	if imageRequest.Prompt == "" {
-		return nil, errors.New("prompt is required")
-	}
-	if strings.Contains(imageRequest.Size, "×") {
-		return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
-	}
-	if imageRequest.N == 0 {
-		imageRequest.N = 1
-	}
-	if imageRequest.Size == "" {
-		imageRequest.Size = "1024x1024"
-	}
-	if imageRequest.Model == "" {
-		imageRequest.Model = "dall-e-2"
-	}
-	// x.ai grok-2-image not support size, quality or style
-	if imageRequest.Size == "empty" {
-		imageRequest.Size = ""
-	}
-
-	// Not "256x256", "512x512", or "1024x1024"
-	if imageRequest.Model == "dall-e-2" || imageRequest.Model == "dall-e" {
-		if imageRequest.Size != "" && imageRequest.Size != "256x256" && imageRequest.Size != "512x512" && imageRequest.Size != "1024x1024" {
-			return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024, dall-e-3 1024x1792 or 1792x1024")
-		}
-	} else if imageRequest.Model == "dall-e-3" {
-		if imageRequest.Size != "" && imageRequest.Size != "1024x1024" && imageRequest.Size != "1024x1792" && imageRequest.Size != "1792x1024" {
-			return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024, dall-e-3 1024x1792 or 1792x1024")
-		}
-		if imageRequest.Quality == "" {
-			imageRequest.Quality = "standard"
-		}
-		//if imageRequest.N != 1 {
-		//	return nil, errors.New("n must be 1")
-		//}
-	}
-	// N should between 1 and 10
-	//if imageRequest.N != 0 && (imageRequest.N < 1 || imageRequest.N > 10) {
-	//	return service.OpenAIErrorWrapper(errors.New("n must be between 1 and 10"), "invalid_field_value", http.StatusBadRequest)
-	//}
 	if setting.ShouldCheckPromptSensitive() {
 		words, err := service.CheckSensitiveInput(imageRequest.Prompt)
 		if err != nil {
@@ -229,6 +191,10 @@ func ImageHelper(c *gin.Context) *dto.OpenAIErrorWithStatusCode {
 		requestBody = bytes.NewBuffer(jsonData)
 	}
 
+	if common.DebugEnabled {
+		println(fmt.Sprintf("image request body: %s", requestBody))
+	}
+
 	statusCodeMappingStr := c.GetString("status_code_mapping")
 
 	resp, err := adaptor.DoRequest(c, relayInfo, requestBody)

relay/relay-text.go

@@ -47,6 +47,20 @@ func getAndValidateTextRequest(c *gin.Context, relayInfo *relaycommon.RelayInfo)
 	if textRequest.Model == "" {
 		return nil, errors.New("model is required")
 	}
+	if textRequest.WebSearchOptions != nil {
+		if textRequest.WebSearchOptions.SearchContextSize != "" {
+			validSizes := map[string]bool{
+				"high":   true,
+				"medium": true,
+				"low":    true,
+			}
+			if !validSizes[textRequest.WebSearchOptions.SearchContextSize] {
+				return nil, errors.New("invalid search_context_size, must be one of: high, medium, low")
+			}
+		} else {
+			textRequest.WebSearchOptions.SearchContextSize = "medium"
+		}
+	}
 	switch relayInfo.RelayMode {
 	case relayconstant.RelayModeCompletions:
 		if textRequest.Prompt == "" {
@@ -76,6 +90,10 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
 
 	// get & validate textRequest 获取并验证文本请求
 	textRequest, err := getAndValidateTextRequest(c, relayInfo)
+	if textRequest.WebSearchOptions != nil {
+		c.Set("chat_completion_web_search_context_size", textRequest.WebSearchOptions.SearchContextSize)
+	}
+
 	if err != nil {
 		common.LogError(c, fmt.Sprintf("getAndValidateTextRequest failed: %s", err.Error()))
 		return service.OpenAIErrorWrapperLocal(err, "invalid_text_request", http.StatusBadRequest)
@@ -370,9 +388,20 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 			dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
 				Mul(decimal.NewFromInt(int64(webSearchTool.CallCount))).
 				Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-			extraContent += fmt.Sprintf("Web Search 调用 %d 次，上下文大小 %s，调用花费 $%s",
+			extraContent += fmt.Sprintf("Web Search 调用 %d 次，上下文大小 %s，调用花费 %s",
 				webSearchTool.CallCount, webSearchTool.SearchContextSize, dWebSearchQuota.String())
 		}
+	} else if strings.HasSuffix(modelName, "search-preview") {
+		// search-preview 模型不支持 response api
+		searchContextSize := ctx.GetString("chat_completion_web_search_context_size")
+		if searchContextSize == "" {
+			searchContextSize = "medium"
+		}
+		webSearchPrice = operation_setting.GetWebSearchPricePerThousand(modelName, searchContextSize)
+		dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
+			Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
+		extraContent += fmt.Sprintf("Web Search 调用 1 次，上下文大小 %s，调用花费 %s",
+			searchContextSize, dWebSearchQuota.String())
 	}
 	// file search tool 计费
 	var dFileSearchQuota decimal.Decimal
@@ -463,10 +492,16 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 		other["image_ratio"] = imageRatio
 		other["image_output"] = imageTokens
 	}
-	if !dWebSearchQuota.IsZero() && relayInfo.ResponsesUsageInfo != nil {
-		if webSearchTool, exists := relayInfo.ResponsesUsageInfo.BuiltInTools[dto.BuildInToolWebSearchPreview]; exists {
+	if !dWebSearchQuota.IsZero() {
+		if relayInfo.ResponsesUsageInfo != nil {
+			if webSearchTool, exists := relayInfo.ResponsesUsageInfo.BuiltInTools[dto.BuildInToolWebSearchPreview]; exists {
+				other["web_search"] = true
+				other["web_search_call_count"] = webSearchTool.CallCount
+				other["web_search_price"] = webSearchPrice
+			}
+		} else if strings.HasSuffix(modelName, "search-preview") {
 			other["web_search"] = true
-			other["web_search_call_count"] = webSearchTool.CallCount
+			other["web_search_call_count"] = 1
 			other["web_search_price"] = webSearchPrice
 		}
 	}

web/src/components/ChannelsTable.js

@@ -888,9 +888,13 @@ const ChannelsTable = () => {
     const localIdSort = localStorage.getItem('id-sort') === 'true';
     const localPageSize =
       parseInt(localStorage.getItem('page-size')) || ITEMS_PER_PAGE;
+    const localEnableTagMode = localStorage.getItem('enable-tag-mode') === 'true';
+    const localEnableBatchDelete = localStorage.getItem('enable-batch-delete') === 'true';
     setIdSort(localIdSort);
     setPageSize(localPageSize);
-    loadChannels(0, localPageSize, localIdSort, enableTagMode)
+    setEnableTagMode(localEnableTagMode);
+    setEnableBatchDelete(localEnableBatchDelete);
+    loadChannels(0, localPageSize, localIdSort, localEnableTagMode)
       .then()
       .catch((reason) => {
         showError(reason);
@@ -1486,10 +1490,12 @@ const ChannelsTable = () => {
             {t('开启批量操作')}
           </Typography.Text>
           <Switch
+            checked={enableBatchDelete}
             label={t('开启批量操作')}
             uncheckedText={t('关')}
             aria-label={t('是否开启批量操作')}
             onChange={(v) => {
+              localStorage.setItem('enable-batch-delete', v + '');
               setEnableBatchDelete(v);
             }}
           />
@@ -1553,6 +1559,7 @@ const ChannelsTable = () => {
             uncheckedText={t('关')}
             aria-label={t('是否启用标签聚合')}
             onChange={(v) => {
+              localStorage.setItem('enable-tag-mode', v + '');
               setEnableTagMode(v);
               loadChannels(0, pageSize, idSort, v);
             }}

web/src/pages/Home/index.js

@@ -158,7 +158,7 @@ const Home = () => {
                   </p>
                   <p>
                     {t('OIDC 身份验证')}：
-                    {statusState?.status?.oidc === true
+                    {statusState?.status?.oidc_enabled === true
                       ? t('已启用')
                       : t('未启用')}
                   </p>

web/src/pages/Playground/Playground.js

@@ -64,8 +64,9 @@ const Playground = () => {
     },
   ];
 
+  const defaultModel = 'gpt-4o-mini';
   const [inputs, setInputs] = useState({
-    model: 'gpt-4o-mini',
+    model: defaultModel,
     group: '',
     max_tokens: 0,
     temperature: 0,
@@ -108,6 +109,11 @@ const Playground = () => {
         value: model,
       }));
       setModels(localModelOptions);
+      // if default model is not in the list, set the first one as default
+      const hasDefault = localModelOptions.some(option => option.value === defaultModel);
+      if (!hasDefault && localModelOptions.length > 0) {
+        setInputs((inputs) => ({ ...inputs, model: localModelOptions[0].value }));
+      }
     } else {
       showError(t(message));
     }