gemini text generation

relay/channel/gemini/adaptor.go

@@ -10,6 +10,7 @@ import (
 	"one-api/dto"
 	"one-api/relay/channel"
 	relaycommon "one-api/relay/common"
+	"one-api/relay/constant"
 	"one-api/service"
 	"one-api/setting/model_setting"
 	"strings"
@@ -165,6 +166,11 @@ func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, request
 }
 
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage any, err *dto.OpenAIErrorWithStatusCode) {
+	if info.RelayMode == constant.RelayModeGemini {
+		err, usage = GeminiTextGenerationHandler(c, resp, info)
+		return usage, err
+	}
+
 	if strings.HasPrefix(info.UpstreamModelName, "imagen") {
 		return GeminiImageHandler(c, resp, info)
 	}

relay/channel/gemini/relay-gemini-native.go

@@ -0,0 +1,77 @@
+package gemini
+
+import (
+	"encoding/json"
+	"io"
+	"net/http"
+	"one-api/common"
+	"one-api/dto"
+	relaycommon "one-api/relay/common"
+	"one-api/service"
+
+	"github.com/gin-gonic/gin"
+)
+
+func GeminiTextGenerationHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {
+	// 读取响应体
+	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
+	}
+
+	if common.DebugEnabled {
+		println(string(responseBody))
+	}
+
+	// 解析为 Gemini 原生响应格式
+	var geminiResponse dto.GeminiTextGenerationResponse
+	err = common.DecodeJson(responseBody, &geminiResponse)
+	if err != nil {
+		return service.OpenAIErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	// 检查是否有候选响应
+	if len(geminiResponse.Candidates) == 0 {
+		return &dto.OpenAIErrorWithStatusCode{
+			Error: dto.OpenAIError{
+				Message: "No candidates returned",
+				Type:    "server_error",
+				Param:   "",
+				Code:    500,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+
+	// 计算使用量（基于 UsageMetadata）
+	usage := dto.Usage{
+		PromptTokens:     geminiResponse.UsageMetadata.PromptTokenCount,
+		CompletionTokens: geminiResponse.UsageMetadata.CandidatesTokenCount,
+		TotalTokens:      geminiResponse.UsageMetadata.TotalTokenCount,
+	}
+
+	// 设置模型版本
+	if geminiResponse.ModelVersion == "" {
+		geminiResponse.ModelVersion = info.UpstreamModelName
+	}
+
+	// 直接返回 Gemini 原生格式的 JSON 响应
+	jsonResponse, err := json.Marshal(geminiResponse)
+	if err != nil {
+		return service.OpenAIErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	// 设置响应头并写入响应
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(jsonResponse)
+	if err != nil {
+		return service.OpenAIErrorWrapper(err, "write_response_failed", http.StatusInternalServerError), nil
+	}
+
+	return nil, &usage
+}

relay/constant/relay_mode.go

@@ -43,6 +43,8 @@ const (
 	RelayModeResponses
 
 	RelayModeRealtime
+
+	RelayModeGemini
 )
 
 func Path2RelayMode(path string) int {
@@ -75,6 +77,8 @@ func Path2RelayMode(path string) int {
 		relayMode = RelayModeRerank
 	} else if strings.HasPrefix(path, "/v1/realtime") {
 		relayMode = RelayModeRealtime
+	} else if strings.HasPrefix(path, "/v1beta/models") {
+		relayMode = RelayModeGemini
 	}
 	return relayMode
 }

relay/relay-gemini.go

@@ -0,0 +1,141 @@
+package relay
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net/http"
+	"one-api/common"
+	"one-api/dto"
+	relaycommon "one-api/relay/common"
+	"one-api/relay/helper"
+	"one-api/service"
+	"one-api/setting"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+)
+
+func getAndValidateGeminiRequest(c *gin.Context) (*dto.GeminiTextGenerationRequest, error) {
+	request := &dto.GeminiTextGenerationRequest{}
+	err := common.UnmarshalBodyReusable(c, request)
+	if err != nil {
+		return nil, err
+	}
+	if len(request.Contents) == 0 {
+		return nil, errors.New("contents is required")
+	}
+	return request, nil
+}
+
+func checkGeminiInputSensitive(textRequest *dto.GeminiTextGenerationRequest, info *relaycommon.RelayInfo) ([]string, error) {
+	var inputTexts []string
+	for _, content := range textRequest.Contents {
+		for _, part := range content.Parts {
+			if part.Text != "" {
+				inputTexts = append(inputTexts, part.Text)
+			}
+		}
+	}
+	if len(inputTexts) == 0 {
+		return nil, nil
+	}
+
+	sensitiveWords, err := service.CheckSensitiveInput(inputTexts)
+	return sensitiveWords, err
+}
+
+func getGeminiInputTokens(req *dto.GeminiTextGenerationRequest, info *relaycommon.RelayInfo) (int, error) {
+	// 计算输入 token 数量
+	var inputTexts []string
+	for _, content := range req.Contents {
+		for _, part := range content.Parts {
+			if part.Text != "" {
+				inputTexts = append(inputTexts, part.Text)
+			}
+		}
+	}
+
+	inputText := strings.Join(inputTexts, "\n")
+	inputTokens, err := service.CountTokenInput(inputText, info.UpstreamModelName)
+	info.PromptTokens = inputTokens
+	return inputTokens, err
+}
+
+func GeminiHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
+	req, err := getAndValidateGeminiRequest(c)
+	if err != nil {
+		common.LogError(c, fmt.Sprintf("getAndValidateGeminiRequest error: %s", err.Error()))
+		return service.OpenAIErrorWrapperLocal(err, "invalid_gemini_request", http.StatusBadRequest)
+	}
+
+	relayInfo := relaycommon.GenRelayInfo(c)
+
+	if setting.ShouldCheckPromptSensitive() {
+		sensitiveWords, err := checkGeminiInputSensitive(req, relayInfo)
+		if err != nil {
+			common.LogWarn(c, fmt.Sprintf("user sensitive words detected: %s", strings.Join(sensitiveWords, ", ")))
+			return service.OpenAIErrorWrapperLocal(err, "check_request_sensitive_error", http.StatusBadRequest)
+		}
+	}
+
+	// model mapped 模型映射
+	err = helper.ModelMappedHelper(c, relayInfo)
+	if err != nil {
+		return service.OpenAIErrorWrapperLocal(err, "model_mapped_error", http.StatusBadRequest)
+	}
+
+	if value, exists := c.Get("prompt_tokens"); exists {
+		promptTokens := value.(int)
+		relayInfo.SetPromptTokens(promptTokens)
+	} else {
+		promptTokens, err := getGeminiInputTokens(req, relayInfo)
+		if err != nil {
+			return service.OpenAIErrorWrapperLocal(err, "count_input_tokens_error", http.StatusBadRequest)
+		}
+		c.Set("prompt_tokens", promptTokens)
+	}
+
+	priceData, err := helper.ModelPriceHelper(c, relayInfo, relayInfo.PromptTokens, req.GenerationConfig.MaxOutputTokens)
+	if err != nil {
+		return service.OpenAIErrorWrapperLocal(err, "model_price_error", http.StatusInternalServerError)
+	}
+
+	// pre consume quota
+	preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
+	if openaiErr != nil {
+		return openaiErr
+	}
+	defer func() {
+		if openaiErr != nil {
+			returnPreConsumedQuota(c, relayInfo, userQuota, preConsumedQuota)
+		}
+	}()
+
+	adaptor := GetAdaptor(relayInfo.ApiType)
+	if adaptor == nil {
+		return service.OpenAIErrorWrapperLocal(fmt.Errorf("invalid api type: %d", relayInfo.ApiType), "invalid_api_type", http.StatusBadRequest)
+	}
+
+	adaptor.Init(relayInfo)
+
+	requestBody, err := json.Marshal(req)
+	if err != nil {
+		return service.OpenAIErrorWrapperLocal(err, "marshal_text_request_failed", http.StatusInternalServerError)
+	}
+
+	resp, err := adaptor.DoRequest(c, relayInfo, bytes.NewReader(requestBody))
+	if err != nil {
+		common.LogError(c, "Do gemini request failed: "+err.Error())
+		return service.OpenAIErrorWrapperLocal(err, "do_request_failed", http.StatusInternalServerError)
+	}
+
+	usage, openaiErr := adaptor.DoResponse(c, resp.(*http.Response), relayInfo)
+	if openaiErr != nil {
+		return openaiErr
+	}
+
+	postConsumeQuota(c, relayInfo, usage.(*dto.Usage), preConsumedQuota, userQuota, priceData, "")
+	return nil
+}