Properly support per-model maxTokens/temperature

pkg/cmd/util/util.go

@@ -19,15 +19,16 @@ import (
 // Prompt prompts the configured the configured model and streams the response
 // to stdout. Returns all model reply messages.
 func Prompt(ctx *lmcli.Context, messages []conversation.Message, callback func(conversation.Message)) (*api.Message, error) {
-	m, _, p, err := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
+	modelConfig, err := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
 	if err != nil {
 		return nil, err
 	}
+	p := modelConfig.Client
 
 	params := provider.RequestParameters{
-		Model:       m,
+		Model:       modelConfig.Model,
-		MaxTokens:   *ctx.Config.Defaults.MaxTokens,
+		MaxTokens:   modelConfig.MaxTokens,
-		Temperature: *ctx.Config.Defaults.Temperature,
+		Temperature: modelConfig.Temperature,
 	}
 
 	system := ctx.DefaultSystemPrompt()
@@ -206,15 +207,16 @@ Example response:
 		},
 	}
 
-	m, _, p, err := ctx.GetModelProvider(
+	modelConfig, err := ctx.GetModelProvider(
 		*ctx.Config.Conversations.TitleGenerationModel, "",
 	)
 	if err != nil {
 		return "", err
 	}
+	p := modelConfig.Client
 
 	requestParams := provider.RequestParameters{
-		Model:     m,
+		Model:     modelConfig.Model,
 		MaxTokens: 25,
 	}
 

pkg/lmcli/lmcli.go

@@ -139,7 +139,35 @@ func (c *Context) DefaultSystemPrompt() string {
 	return c.Config.Defaults.SystemPrompt
 }
 
-func (c *Context) GetModelProvider(model string, provider string) (string, string, provider.ChatCompletionProvider, error) {
+type ModelConfig struct {
+	Provider	string
+	Client     provider.ChatCompletionProvider
+	Model       string
+	MaxTokens   int
+	Temperature float32
+}
+
+func (c *Context) fillModelConfig(cfg *ModelConfig, m Model) *ModelConfig {
+	// Set model name
+	cfg.Model = m.Name
+
+	// Set max tokens
+	if m.MaxTokens == nil {
+		cfg.MaxTokens = *c.Config.Defaults.MaxTokens
+	} else {
+		cfg.MaxTokens = *m.MaxTokens
+	}
+
+	// Set temperature
+	if m.Temperature == nil {
+		cfg.Temperature = *c.Config.Defaults.Temperature
+	} else {
+		cfg.Temperature = *m.Temperature
+	}
+	return cfg
+}
+
+func (c *Context) GetModelProvider(model string, provider string) (*ModelConfig, error) {
 	parts := strings.Split(model, "@")
 
 	if provider == "" && len(parts) > 1 {
@@ -158,6 +186,7 @@ func (c *Context) GetModelProvider(model string, provider string) (string, strin
 		}
 
 		for _, m := range p.Models() {
+			var cfg *ModelConfig
 			if m.Name == model {
 				switch p.Kind {
 				case "anthropic":
@@ -165,44 +194,56 @@ func (c *Context) GetModelProvider(model string, provider string) (string, strin
 					if p.BaseURL != "" {
 						url = p.BaseURL
 					}
-					return model, name, &anthropic.AnthropicClient{
+					cfg = &ModelConfig{
+						Client: &anthropic.AnthropicClient{
 							BaseURL: url,
 							APIKey:  p.APIKey,
-					}, nil
+						},
+					}
+					return c.fillModelConfig(cfg, m), nil
 				case "google":
 					url := "https://generativelanguage.googleapis.com"
 					if p.BaseURL != "" {
 						url = p.BaseURL
 					}
-					return model, name, &google.Client{
+					cfg := &ModelConfig{
+						Client: &google.Client{
 							BaseURL: url,
 							APIKey:  p.APIKey,
-					}, nil
+						},
+					}
+					return c.fillModelConfig(cfg, m), nil
 				case "ollama":
 					url := "http://localhost:11434/api"
 					if p.BaseURL != "" {
 						url = p.BaseURL
 					}
-					return model, name, &ollama.OllamaClient{
+					cfg := &ModelConfig{
+						Client: &ollama.OllamaClient{
 							BaseURL: url,
-					}, nil
+						},
+					}
+					return c.fillModelConfig(cfg, m), nil
 				case "openai":
 					url := "https://api.openai.com"
 					if p.BaseURL != "" {
 						url = p.BaseURL
 					}
-					return model, name, &openai.OpenAIClient{
+					cfg := &ModelConfig{
+						Client: &openai.OpenAIClient{
 							BaseURL: url,
 							APIKey:  p.APIKey,
 							Headers: p.Headers,
-					}, nil
+						},
+					}
+					return c.fillModelConfig(cfg, m), nil
 				default:
-					return "", "", nil, fmt.Errorf("unknown provider kind: %s", p.Kind)
+					return nil, fmt.Errorf("unknown provider kind: %s", p.Kind)
 				}
 			}
 		}
 	}
-	return "", "", nil, fmt.Errorf("unknown model: %s", model)
+	return nil, fmt.Errorf("unknown model: %s", model)
 }
 
 func Fatal(format string, args ...any) {

pkg/tui/model/model.go

@@ -39,9 +39,9 @@ func NewAppModel(ctx *lmcli.Context, initialConversation *conversation.Conversat
 
 	}
 
-	model, provider, _, _ := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
+	modelConfig, _ := ctx.GetModelProvider(*ctx.Config.Defaults.Model, "")
-	app.Model = model
+	app.Model = modelConfig.Model
-	app.ProviderName = provider
+	app.ProviderName = modelConfig.Provider
 	app.Agent = ctx.GetAgent(ctx.Config.Defaults.Agent)
 	return app
 }
@@ -257,15 +257,16 @@ func (a *AppModel) Prompt(
 	chatReplyChunks chan provider.Chunk,
 	stopSignal chan struct{},
 ) (*conversation.Message, error) {
-	model, _, p, err := a.Ctx.GetModelProvider(a.Model, a.ProviderName)
+	modelConfig, err := a.Ctx.GetModelProvider(a.Model, a.ProviderName)
 	if err != nil {
 		return nil, err
 	}
+	p := modelConfig.Client
 
 	params := provider.RequestParameters{
-		Model:       model,
+		Model:       modelConfig.Model,
-		MaxTokens:   *a.Ctx.Config.Defaults.MaxTokens,
+		MaxTokens:   modelConfig.MaxTokens,
-		Temperature: *a.Ctx.Config.Defaults.Temperature,
+		Temperature: modelConfig.Temperature,
 	}
 
 	if a.Agent != nil {