M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

//go:build !cgo // Package localllm - Generation API Stub // // This file provides stub types for text generation when CGO is not available. // The actual CGO implementation is in llama.go. // // Generation support enables NornicDB to run reasoning SLMs alongside // embedding models for cognitive database capabilities. // // When CGO is enabled, use LoadGenerationModel to load a reasoning model: // // opts := localllm.DefaultGenerationOptions("/models/qwen2.5-0.5b.gguf") // model, err := localllm.LoadGenerationModel(opts) // if err != nil { // log.Fatal(err) // } // defer model.Close() // // response, err := model.Generate(ctx, prompt, DefaultGenerateParams()) package localllm import ( "context" "fmt" ) // GenerationModel wraps a GGUF model for text generation (reasoning tasks). // This is a stub - actual implementation requires CGO. type GenerationModel struct { modelPath string maxContext int } // GenerateParams configures text generation behavior. type GenerateParams struct { MaxTokens int // Maximum tokens to generate (default: 256) Temperature float32 // Sampling temperature (default: 0.1 for structured output) TopP float32 // Nucleus sampling (default: 0.9) TopK int // Top-K sampling (default: 40) StopTokens []string // Stop generation on these strings } // DefaultGenerateParams returns parameters optimized for structured JSON output. func DefaultGenerateParams() GenerateParams { return GenerateParams{ MaxTokens: 256, Temperature: 0.1, TopP: 0.9, TopK: 40, StopTokens: []string{"<|im_end|>", "<|endoftext|>", "</s>"}, } } // GenerationOptions configures model loading for text generation. type GenerationOptions struct { ModelPath string ContextSize int // Max context window (default: 2048) BatchSize int Threads int GPULayers int } // DefaultGenerationOptions returns options optimized for reasoning SLMs. func DefaultGenerationOptions(modelPath string) GenerationOptions { return GenerationOptions{ ModelPath: modelPath, ContextSize: 2048, BatchSize: 512, Threads: 4, GPULayers: -1, } } // LoadGenerationModel is a stub - requires CGO build. func LoadGenerationModel(opts GenerationOptions) (*GenerationModel, error) { return nil, fmt.Errorf("generation model requires CGO build (use -tags cgo)") } // Generate is a stub - requires CGO build. func (m *GenerationModel) Generate(ctx context.Context, prompt string, params GenerateParams) (string, error) { return "", fmt.Errorf("generation requires CGO build") } // GenerateStream is a stub - requires CGO build. func (m *GenerationModel) GenerateStream(ctx context.Context, prompt string, params GenerateParams, callback func(token string) error) error { return fmt.Errorf("generation requires CGO build") } // Close is a stub. func (m *GenerationModel) Close() error { return nil } // ModelPath returns the path to the loaded model file. func (m *GenerationModel) ModelPath() string { return m.modelPath }