feat: support gemini embeddings (text-embedding-004,embedding-001) (#1475)

* Refactor Gemini Adaptor to Support Embeddings

* Add new models to ModelList

relay/adaptor/gemini/main.go

@@ -134,6 +134,29 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 	return &geminiRequest
 }
 
+func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) *BatchEmbeddingRequest {
+	inputs := request.ParseInput()
+	requests := make([]EmbeddingRequest, len(inputs))
+	model := fmt.Sprintf("models/%s", request.Model)
+
+	for i, input := range inputs {
+		requests[i] = EmbeddingRequest{
+			Model: model,
+			Content: ChatContent{
+				Parts: []Part{
+					{
+						Text: input,
+					},
+				},
+			},
+		}
+	}
+
+	return &BatchEmbeddingRequest{
+		Requests: requests,
+	}
+}
+
 type ChatResponse struct {
 	Candidates     []ChatCandidate    `json:"candidates"`
 	PromptFeedback ChatPromptFeedback `json:"promptFeedback"`
@@ -230,6 +253,23 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *ChatResponse) *openai.ChatC
 	return &response
 }
 
+func embeddingResponseGemini2OpenAI(response *EmbeddingResponse) *openai.EmbeddingResponse {
+	openAIEmbeddingResponse := openai.EmbeddingResponse{
+		Object: "list",
+		Data:   make([]openai.EmbeddingResponseItem, 0, len(response.Embeddings)),
+		Model:  "gemini-embedding",
+		Usage:  model.Usage{TotalTokens: 0},
+	}
+	for _, item := range response.Embeddings {
+		openAIEmbeddingResponse.Data = append(openAIEmbeddingResponse.Data, openai.EmbeddingResponseItem{
+			Object:    `embedding`,
+			Index:     0,
+			Embedding: item.Values,
+		})
+	}
+	return &openAIEmbeddingResponse
+}
+
 func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, string) {
 	responseText := ""
 	scanner := bufio.NewScanner(resp.Body)
@@ -337,3 +377,39 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	_, err = c.Writer.Write(jsonResponse)
 	return nil, &usage
 }
+
+func EmbeddingHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	var geminiEmbeddingResponse EmbeddingResponse
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = json.Unmarshal(responseBody, &geminiEmbeddingResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	if geminiEmbeddingResponse.Error != nil {
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: geminiEmbeddingResponse.Error.Message,
+				Type:    "gemini_error",
+				Param:   "",
+				Code:    geminiEmbeddingResponse.Error.Code,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+	fullTextResponse := embeddingResponseGemini2OpenAI(&geminiEmbeddingResponse)
+	jsonResponse, err := json.Marshal(fullTextResponse)
+	if err != nil {
+		return openai.ErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError), nil
+	}
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(jsonResponse)
+	return nil, &fullTextResponse.Usage
+}