Method QueryText

model/volcengine.go:239–337

(question string, writer io.Writer, history []*RawMessage, prompt string, knowledgeMessages []*RawMessage, toolSession *ToolSession, lang string)

}

func (p *VolcengineModelProvider) QueryText(question string, writer io.Writer, history []*RawMessage, prompt string, knowledgeMessages []*RawMessage, toolSession *ToolSession, lang string) (*ModelResult, error) {
	ctx := context.Background()
	flusher, ok := writer.(http.Flusher)
	if !ok {
		return nil, fmt.Errorf(i18n.Translate(lang, "model:writer does not implement http.Flusher"))
	}

	client := arkruntime.NewClientWithApiKey(p.apiKey)

	messages := p.buildMessages(question, history, prompt, knowledgeMessages)

	request := model.CreateChatCompletionRequest{
		Model:         p.subType,
		Messages:      messages,
		Temperature:   &p.temperature,
		TopP:          &p.topP,
		StreamOptions: &model.StreamOptions{IncludeUsage: true},
	}

	thinkingType := model.ThinkingTypeDisabled
	if p.enableThinking {
		thinkingType = model.ThinkingTypeEnabled
	}
	request.Thinking = &model.Thinking{Type: thinkingType}

	flushData := func(data, eventType string) error {
		if _, err := fmt.Fprintf(writer, "event: %s\ndata: %s\n\n", eventType, data); err != nil {
			return err
		}
		flusher.Flush()
		return nil
	}

	stream, err := client.CreateChatCompletionStream(ctx, request)
	if err != nil {
		logs.Error("stream chat error: %v", err)
		return nil, err
	}
	defer stream.Close()

	modelResult := newModelResult(0, 0, 0)
	var reasoningData strings.Builder
	isLeadingReturn := true

	for {
		response, recvErr := stream.Recv()

		if response.Usage != nil {
			modelResult.PromptTokenCount += response.Usage.PromptTokens
			modelResult.ResponseTokenCount += response.Usage.CompletionTokens
			modelResult.TotalTokenCount += response.Usage.TotalTokens
		}

		if recvErr != nil {
			if recvErr == io.EOF {
				break
			}
			return nil, recvErr