refactor(llm): simplify OllamaLLMProvider to embedding-only, remove chat model support

Author: razvan

pkg/llm/ollama.go

@@ -12,20 +12,16 @@ import (
 	"github.com/tmc/langchaingo/llms/ollama"
 )
 
-// OllamaLLMProvider implements Provider interface for Ollama
+// OllamaLLMProvider implements Provider interface for Ollama (embedding-only).
 type OllamaLLMProvider struct {
-	chatModel  llms.Model
 	embedModel llms.Model
-	chatName   string
 	embedName  string
-	config     config.LLMConfig
 	cachedDim  uint64
 	dimOnce    sync.Once
 }
 
-// NewOllamaLLMProvider creates a new Ollama provider with separate chat and embedding models
+// NewOllamaLLMProvider creates a new Ollama provider configured for embedding only.
 func NewOllamaLLMProvider(cfg config.LLMConfig) (*OllamaLLMProvider, error) {
-	// Server URL
 	baseURL := cfg.OllamaBaseURL
 	if baseURL == "" {
 		baseURL = cfg.BaseURL
@@ -34,68 +30,40 @@ func NewOllamaLLMProvider(cfg config.LLMConfig) (*OllamaLLMProvider, error) {
 		baseURL = "http://localhost:11434"
 	}
 
-	// Chat model
-	chatModelName := cfg.OllamaModel
-	if chatModelName == "" {
-		chatModelName = cfg.Model
-	}
-
-	// Embedding model
 	embedModelName := cfg.OllamaEmbed
 	if embedModelName == "" {
 		embedModelName = cfg.EmbedModel
 	}
-
-	// Fallback logic
-	if chatModelName == "" && embedModelName != "" {
-		chatModelName = embedModelName
+	// Accept OllamaModel / Model as fallback for backward-compat
+	if embedModelName == "" {
+		embedModelName = cfg.OllamaModel
 	}
-	if embedModelName == "" && chatModelName != "" {
-		embedModelName = chatModelName
+	if embedModelName == "" {
+		embedModelName = cfg.Model
 	}
-
-	if chatModelName == "" {
-		return nil, fmt.Errorf("ollama model is required (set ollama_model or ollama_embed)")
+	if embedModelName == "" {
+		return nil, fmt.Errorf("ollama model is required (set ollama_embed in config)")
 	}
 
-	// Create chat client
-	chatClient, err := ollama.New(
+	embedClient, err := ollama.New(
 		ollama.WithServerURL(baseURL),
-		ollama.WithModel(chatModelName),
+		ollama.WithModel(embedModelName),
 	)
 	if err != nil {
-		return nil, fmt.Errorf("failed to create Ollama chat client: %w", err)
+		return nil, fmt.Errorf("failed to create Ollama embedding client: %w", err)
 	}
 
-	// Create embedding client (separate if different model)
-	var embedClient llms.Model
-	if embedModelName != chatModelName {
-		embedClient, err = ollama.New(
-			ollama.WithServerURL(baseURL),
-			ollama.WithModel(embedModelName),
-		)
-		if err != nil {
-			return nil, fmt.Errorf("failed to create Ollama embedding client: %w", err)
-		}
-		log.Printf("🎯 Ollama: chat=%s, embed=%s (dual-model)", chatModelName, embedModelName)
-	} else {
-		embedClient = chatClient
-		log.Printf("🎯 Ollama: model=%s (single-model)", chatModelName)
-	}
+	log.Printf("🎯 Ollama: embed=%s", embedModelName)
 
 	return &OllamaLLMProvider{
-		chatModel:  chatClient,
 		embedModel: embedClient,
-		chatName:   chatModelName,
 		embedName:  embedModelName,
-		config:     cfg,
 	}, nil
 }
 
-// Generate generates text using Ollama chat model
-func (p *OllamaLLMProvider) Generate(ctx context.Context, prompt string, opts ...GenerateOption) (string, error) {
-	lcOpts := p.convertOptions(opts...)
-	return llms.GenerateFromSinglePrompt(ctx, p.chatModel, prompt, lcOpts...)
+// Generate is not supported; this provider is embedding-only.
+func (p *OllamaLLMProvider) Generate(_ context.Context, _ string, _ ...GenerateOption) (string, error) {
+	return "", fmt.Errorf("text generation not supported: provider is configured for embedding only")
 }
 
 // GetEmbeddingDimension returns the dimension of the embedding model
@@ -192,33 +160,15 @@ func (p *OllamaLLMProvider) lookupHardcodedDimension() uint64 {
 	}
 }
 
-// GenerateStream generates streaming text using Ollama chat model
-func (p *OllamaLLMProvider) GenerateStream(ctx context.Context, prompt string, opts ...GenerateOption) (<-chan string, <-chan error) {
+// GenerateStream is not supported; this provider is embedding-only.
+func (p *OllamaLLMProvider) GenerateStream(_ context.Context, _ string, _ ...GenerateOption) (<-chan string, <-chan error) {
 	textChan := make(chan string)
 	errChan := make(chan error, 1)
-
 	go func() {
 		defer close(textChan)
 		defer close(errChan)
-
-		streamFunc := func(ctx context.Context, chunk []byte) error {
-			select {
-			case textChan <- string(chunk):
-				return nil
-			case <-ctx.Done():
-				return ctx.Err()
-			}
-		}
-
-		lcOpts := p.convertOptions(opts...)
-		lcOpts = append(lcOpts, llms.WithStreamingFunc(streamFunc))
-
-		_, err := llms.GenerateFromSinglePrompt(ctx, p.chatModel, prompt, lcOpts...)
-		if err != nil {
-			errChan <- err
-		}
+		errChan <- fmt.Errorf("text generation not supported: provider is configured for embedding only")
 	}()
-
 	return textChan, errChan
 }
 
@@ -253,39 +203,3 @@ func (p *OllamaLLMProvider) Embed(ctx context.Context, text string) ([]float64,
 func (p *OllamaLLMProvider) Name() string {
 	return "ollama"
 }
-
-// convertOptions converts GenerateOption to langchaingo CallOption
-func (p *OllamaLLMProvider) convertOptions(opts ...GenerateOption) []llms.CallOption {
-	genOpts := &GenerateOptions{}
-	for _, opt := range opts {
-		opt(genOpts)
-	}
-
-	var lcOpts []llms.CallOption
-
-	if genOpts.Temperature != 0 {
-		lcOpts = append(lcOpts, llms.WithTemperature(genOpts.Temperature))
-	}
-	if genOpts.MaxTokens != 0 {
-		lcOpts = append(lcOpts, llms.WithMaxTokens(genOpts.MaxTokens))
-	}
-	if genOpts.TopP != 0 {
-		lcOpts = append(lcOpts, llms.WithTopP(genOpts.TopP))
-	}
-	if genOpts.TopK != 0 {
-		lcOpts = append(lcOpts, llms.WithTopK(genOpts.TopK))
-	}
-	if len(genOpts.StopSequences) > 0 {
-		lcOpts = append(lcOpts, llms.WithStopWords(genOpts.StopSequences))
-	}
-
-	// Apply config defaults
-	if genOpts.Temperature == 0 && p.config.Temperature != 0 {
-		lcOpts = append(lcOpts, llms.WithTemperature(p.config.Temperature))
-	}
-	if genOpts.MaxTokens == 0 && p.config.MaxTokens != 0 {
-		lcOpts = append(lcOpts, llms.WithMaxTokens(p.config.MaxTokens))
-	}
-
-	return lcOpts
-}

pkg/llm/provider_test.go

@@ -51,7 +51,7 @@ func TestNewProvider_OllamaMissingModel(t *testing.T) {
 	if err == nil {
 		t.Fatalf("expected error when ollama model is missing, got nil")
 	}
-	if !strings.Contains(err.Error(), "ollama chat model is required") {
+	if !strings.Contains(err.Error(), "ollama model is required") { //nolint: keep generic check
 		t.Errorf("unexpected error: %v", err)
 	}
 	if p != nil {
@@ -62,7 +62,7 @@ func TestNewProvider_OllamaMissingModel(t *testing.T) {
 func TestNewProvider_DefaultOllama(t *testing.T) {
 	cfg := &config.LLMConfig{
 		Provider:      "", // implicit ollama
-		OllamaModel:   "dummy-model",
+		OllamaEmbed:   "dummy-model",
 		OllamaBaseURL: "http://localhost:11434",
 	}