Module training 

SONA Training System

Templated training pipelines for specialized model adaptation.

Overview

The training module provides:

• Training Templates: Pre-configured training setups for common use cases
• Agent Factory: Create and manage multiple specialized agents
• Training Pipelines: Structured workflows for different verticals
• Federated Learning: Distributed training across ephemeral agents
• Metrics & Results: Comprehensive training analytics

Quick Start

use ruvector_sona::training::{TrainingTemplate, AgentFactory, TrainingPipeline};

// Use a preset template
let template = TrainingTemplate::code_agent();
let pipeline = TrainingPipeline::from_template(template);

// Train on examples
for example in examples {
    pipeline.add_example(example);
}
let results = pipeline.train()?;

Federated Learning

use ruvector_sona::training::{EphemeralAgent, FederatedCoordinator};

// Create coordinator
let mut coordinator = FederatedCoordinator::default_coordinator("main", 3072);

// Ephemeral agents process tasks
let mut agent = EphemeralAgent::default_federated("agent-1", 3072);
agent.process_trajectory(embedding, activations, quality, route, context);

// Export state before termination
let export = agent.export_state();
coordinator.aggregate(export);

Structs

AgentContribution

Agent contribution record

AgentExport

Exported state from an ephemeral agent

AgentExportStats

Agent export statistics

AgentFactory

Factory for creating and managing agents

AgentHandle

Handle to a managed agent

AgentStats

Agent statistics

AggregationResult

Result of aggregating an agent export

BatchConfig

Batch configuration for training

CoordinatorStats

Coordinator statistics

EphemeralAgent

Ephemeral agent for federated learning

EpochStats

Epoch statistics

FederatedCoordinator

Federated learning coordinator

ManagedAgent

Managed agent with engine and metadata

PerformanceMetrics

Performance metrics

QualityMetrics

Quality metrics summary

SharedAgentFactory

Thread-safe agent factory wrapper

SimpleExample

Simple training example implementation

TrainingExample

Training example with all data needed for learning

TrainingMetrics

Training metrics collection

TrainingPipeline

Training pipeline for structured training workflows

TrainingResult

Training result summary

TrainingTemplate

Training template with full configuration

TrajectoryExport

Single trajectory export

VerticalConfig

Vertical-specific configuration

Enums

AgentType

Agent specialization types

DataSizeHint

Hint about training data size

FederatedTopology

Federated learning topology

PipelineStage

Pipeline stage for tracking progress

TaskDomain

Task domain for training focus

TemplatePreset

Template preset for quick configuration

TrainingMethod

Training method configuration

Traits

FactoryTrainingExample

Trait for training examples

TrainingCallback

Callback trait for training events