Training Services Documentation

TrainingRunService

Manages the execution and lifecycle of training runs, handling the core training workflow from initialization to completion.

Key Methods

CreateTrainingRunWithBaseModel

public async Task<ResultDTO<TrainingRunDTO>> CreateTrainingRunWithBaseModel(TrainingRunDTO inputTrainingRunDTO)

• Creates a new training run with transfer learning setup
• Validates base model and dataset
• Returns training run DTO with configuration

ExecuteTrainingRunProcess

public async Task<ResultDTO> ExecuteTrainingRunProcess(TrainingRunDTO trainingRunDTO)

• Handles main training execution
• Manages dataset preparation
• Generates configurations
• Executes MMDetection training
• Creates trained model entity

GenerateTrainingRunConfigFile

public async Task<ResultDTO<string>> GenerateTrainingRunConfigFile(
    Guid trainingRunId, 
    DatasetDTO datasetDTO, 
    TrainedModelDTO baseTrainedModelDTO, 
    int? numEpochs = null, 
    int? numFrozenStages = null, 
    int? numBatchSize = null)

• Generates MMDetection configuration
• Sets up training parameters
• Configures model architecture
• Returns path to config file

StartTrainingRun

public async Task<ResultDTO> StartTrainingRun(Guid trainingRunId)

• Initiates training process
• Executes MMDetection commands
• Monitors training progress

TrainedModelService

Manages trained model entities, handling model metadata, storage, and lifecycle.

Key Methods

GetTrainedModelById

public async Task<ResultDTO<TrainedModelDTO>> GetTrainedModelById(Guid id, bool track = false)

• Retrieves model information
• Includes related entities
• Returns model DTO

GetPublishedTrainedModelsIncludingTrainRuns

public async Task<ResultDTO<List<TrainedModelDTO>>> GetPublishedTrainedModelsIncludingTrainRuns()

• Retrieves published models
• Includes training history
• Returns list of model DTOs

GetBestEpochForTrainedModelById

public async Task<ResultDTO<TrainingRunResultsDTO>> GetBestEpochForTrainedModelById(Guid trainedModelId)

• Analyzes training metrics
• Identifies best performing epoch
• Returns training results

Service Integration

Training Workflow

1. Initialization

   var trainingRunDTO = await _trainingRunService.CreateTrainingRunWithBaseModel(inputDTO);
   

2. Configuration

   var configPath = await _trainingRunService.GenerateTrainingRunConfigFile(
       trainingRunDTO.Id,
       datasetDTO,
       baseModelDTO
   );
   

3. Execution

   await _trainingRunService.StartTrainingRun(trainingRunDTO.Id);
   

4. Model Creation

   var modelId = await _trainingRunService.CreateTrainedModelByTrainingRunId(trainingRunDTO.Id);
   

Error Handling

• Services use ResultDTO pattern
• Comprehensive error logging
• Automatic cleanup on failure
• Status updates for monitoring

Integration Points

• MMDetection framework
• Dataset management
• File system operations
• Background job processing