refactor: add LLMServiceTypeDoubaoVL

uixt/ai/ai.go

@@ -95,13 +95,16 @@ func GetModelConfig(modelType option.LLMServiceType) (*ModelConfig, error) {
 			"env %s missed", EnvModelName)
 	}
 
-	temperature := float32(0.7)
+	// https://www.volcengine.com/docs/82379/1536429
+	temperature := float32(0)
+	topP := float32(0.7)
 	modelConfig := &openai.ChatModelConfig{
 		BaseURL:     openaiBaseURL,
 		APIKey:      openaiAPIKey,
 		Model:       modelName,
 		Timeout:     defaultTimeout,
 		Temperature: &temperature,
+		TopP:        &topP,
 	}
 
 	// log config info

uixt/ai/asserter.go

@@ -55,7 +55,7 @@ func NewAsserter(ctx context.Context, modelConfig *ModelConfig) (*Asserter, erro
 
 	if modelConfig.ModelType == option.LLMServiceTypeUITARS {
 		asserter.systemPrompt += "\n\n" + uiTarsAssertionResponseFormat
-	} else if modelConfig.ModelType == option.LLMServiceTypeGPT {
+	} else if modelConfig.ModelType == option.LLMServiceTypeDoubaoVL {
 		// define output format
 		type OutputFormat struct {
 			Thought string `json:"thought"`

uixt/ai/planner_parser.go

@@ -27,7 +27,7 @@ func NewLLMContentParser(modelType option.LLMServiceType) LLMContentParser {
 		return &UITARSContentParser{
 			systemPrompt: uiTarsPlanningPrompt,
 		}
-	case option.LLMServiceTypeGPT:
+	case option.LLMServiceTypeDoubaoVL:
 		return &JSONContentParser{
 			systemPrompt: defaultPlanningResponseJsonFormat,
 		}

uixt/ai/planner_prompts.go

@@ -30,7 +30,14 @@ finished(content='xxx') # Use escape characters \\', \\", and \\n in content par
 `
 
 // system prompt for JSONContentParser
-const defaultPlanningResponseJsonFormat = `You are a versatile professional in software UI automation.`
+const defaultPlanningResponseJsonFormat = `You are a versatile professional in software UI automation.
+
+## Output Format
+` + "```" + `
+Thought: ...
+Action: ...
+` + "```" + `
+`
 
 const defaultPlanningResponseStringFormat = `
 You are a helpful assistant.