Training and Inference

Training Through Extension DI

When config.Training.Enabled is true, the extension registers:

• ai.ModelTrainer
• ai.DataManager
• ai.PipelineManager

Enable Training Services

ext := ai.NewExtension(
    ai.WithTrainingConfig(ai.TrainingConfiguration{
        Enabled:           true,
        CheckpointPath:    "./checkpoints",
        ModelPath:         "./models",
        DataPath:          "./data",
        MaxConcurrentJobs: 5,
    }),
)

Resolve Services by Type

trainer, err := forge.InjectType[ai.ModelTrainer](app.Container())
dataMgr, err := forge.InjectType[ai.DataManager](app.Container())
pipelineMgr, err := forge.InjectType[ai.PipelineManager](app.Container())

Inference Package Usage

extensions/ai/inference is a standalone package with its own lifecycle.

The extension config has InferenceConfiguration, but the extension does not currently auto-create/start InferenceEngine from that block.

Typical Direct Usage

engine, err := inference.NewInferenceEngine(inference.InferenceConfig{
    Workers:        4,
    BatchSize:      16,
    BatchTimeout:   100 * time.Millisecond,
    EnableBatching: true,
    EnableCaching:  true,
    EnableScaling:  true,
})
if err != nil {
    return err
}

if err := engine.Start(ctx); err != nil {
    return err
}
defer engine.Stop(ctx)

Middleware Package Usage

extensions/ai/middleware provides independent middleware components such as:

• intelligent rate limiting
• anomaly detection
• personalization
• response optimization
• adaptive load balancing
• security scanning

Each middleware has its own constructor and Initialize/Process flow.

Monitoring Package Usage

extensions/ai/monitoring provides independent monitoring systems:

• AIHealthMonitor
• AIMetricsCollector
• AIAlertManager
• AIDashboard

These are not auto-wired by ai.Extension. Instantiate and run them explicitly.

Practical Guidance

• Use extension DI for shared core services (LLM manager, stores, agents, training interfaces).
• Instantiate inference/middleware/monitoring modules explicitly where you need lifecycle control.
• Keep startup ownership clear: extension bootstraps core AI graph; app code bootstraps advanced pipelines.