Method invokeNonStream

internal/adapters/vllm/vllm.go:65–142

(ctx context.Context, text, model string, streamRequested bool)

}

func (a *Adapter) invokeNonStream(ctx context.Context, text, model string, streamRequested bool) (types.BackendResponse, error) {
	t0 := time.Now()
	payload := map[string]any{
		"model": model,
		"messages": []map[string]string{
			{"role": "user", "content": text},
		},
		"stream": false,
	}

	raw, err := json.Marshal(payload)
	if err != nil {
		return types.BackendResponse{}, err
	}

	chatURL := a.chatCompletionsURL()
	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, chatURL, bytes.NewReader(raw))
	if err != nil {
		return types.BackendResponse{}, err
	}
	httpReq.Header.Set("Content-Type", "application/json")
	a.applyHeaders(httpReq)

	resp, err := a.httpClient.Do(httpReq)
	if err != nil {
		if errors.Is(err, context.DeadlineExceeded) {
			return types.BackendResponse{}, upstream.New(upstream.KindTimeout, err.Error())
		}
		return types.BackendResponse{}, err
	}
	defer resp.Body.Close()
	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
		if resp.StatusCode >= 500 {
			return types.BackendResponse{}, upstream.New(upstream.KindUpstream5xx, fmt.Sprintf("openai-compatible invoke failed status=%d", resp.StatusCode))
		}
		return types.BackendResponse{}, upstream.New(upstream.KindUpstream4xx, fmt.Sprintf("openai-compatible invoke failed status=%d", resp.StatusCode))
	}

	var parsed struct {
		Choices []struct {
			Message struct {
				Content string `json:"content"`
			} `json:"message"`
			Text string `json:"text"`
		} `json:"choices"`
	}
	if err := json.NewDecoder(resp.Body).Decode(&parsed); err != nil {
		return types.BackendResponse{}, upstream.New(upstream.KindBackendError, err.Error())
	}

	output := ""
	if len(parsed.Choices) > 0 {
		if parsed.Choices[0].Message.Content != "" {
			output = parsed.Choices[0].Message.Content
		} else {
			output = parsed.Choices[0].Text
		}
	}