Manage Training Projects

Effective training project management keeps your VLM development environment organized and focused on active work. Datature Vi provides comprehensive tools to rename, delete, and organize training projects as your machine learning initiatives evolve.

This guide covers all aspects of training project management, from organizing workflows and runs to cleaning up completed or abandoned projects.

💡

Complete VLM training workflow

Create training project → Manage training projects (you are here) → Create workflow → Start training → Evaluate model

What are training projects?

Training projects are dedicated workspaces that organize all components of your VLM development:

• Workflows — Reusable training configurations with system prompts, dataset settings, and model parameters
• Runs — Individual training sessions that execute workflows and produce trained models
• Models — Trained artifacts ready for evaluation and deployment
• History — Complete audit trail of experiments, metrics, and results

Each project provides isolated organization for a specific machine learning task or use case.

📘

Project organization benefits

• Logical separation — Keep different tasks (defect detection, quality VQA) in separate projects
• Team collaboration — Share projects with team members for coordinated development
• Resource tracking — Monitor Compute Credit usage per project
• Experiment management — Maintain clear history of all training attempts

Core project operations

Datature Vi provides essential training project management operations:

Quick reference

Common training project management tasks:

Renaming training projects

Keep your training environment organized with descriptive project names that clearly communicate purpose.

When to rename

• Improved clarity — Make project purpose obvious to team members
• Scope evolution — Update names as project objectives change
• Naming standards — Apply consistent conventions across organization
• From defaults — Rename generic project names to meaningful identifiers

Key features

• Safe operation — Project ID remains unchanged; all references continue working
• No downtime — Active training runs and workflows are unaffected
• Instant updates — Name changes appear immediately across platform
• Flexible naming — Use any naming convention that works for your team

Important limitations

⚠️

Project localization cannot be changed

While you can rename a project, its geographic localization (data processing region) is permanent. If you need a different localization, you must create a new project.

Learn how to rename training projects →

Deleting training projects

Permanently remove completed, test, or abandoned training projects to maintain an organized workspace.

When to delete

• Completed projects — Remove projects after successful deployment to production
• Test projects — Clean up experimental or prototype projects
• Failed initiatives — Remove abandoned projects that didn't produce viable models
• Duplicate projects — Delete accidentally created duplicates
• Workspace cleanup — Maintain focused environment with only active projects

Safety measures

Before deleting a training project:

1. Download trained models — Export any models you want to keep
2. Check dependencies — Verify no deployments depend on project models
3. Inform team — Notify collaborators about planned deletion
4. Review carefully — Ensure you're deleting the correct project

🔒

Deletion is permanent and irreversible

• All workflows are permanently deleted
• All training runs and history are removed
• All trained models are permanently deleted
• Project ID becomes invalid
• Cannot be recovered through any means

Always download trained models before deletion if you might need them later.

Learn how to delete training projects →

Managing workflows

Workflows are reusable training configurations that define how your VLM learns from data. Effective workflow management enables systematic experimentation and optimization.

Workflow operations

Common workflow scenarios

Learn about workflow management →

Managing training runs

Training runs are individual training sessions that execute workflows. Active run management ensures efficient use of Compute Credits and helps you track experiments.

Run operations

Run lifecycle

1. Queued — Run is scheduled and waiting for GPU resources
2. Initializing — Environment setup and data preparation
3. Training — Active training in progress with live metrics
4. Evaluating — Post-training evaluation on validation set
5. Completed — Training finished successfully
6. Failed — Training encountered an error
7. Cancelled — Manually stopped by user

Learn about run management →

Best practices for project management

Project organization strategies

Organize your training projects systematically as your VLM development scales.

Strategy 1: Task-based organization

Create separate projects for distinct machine learning tasks:

PCB-Component-Detection
PCB-Defect-Classification
PCB-Quality-VQA
PCB-Assembly-Verification

Best for: Multi-task applications requiring different model architectures or datasets

Strategy 2: Stage-based organization

Separate projects by development stage:

PCB-Detection-DEV (Experimentation)
PCB-Detection-STAGING (Pre-production testing)
PCB-Detection-PROD (Production models)

Best for: Structured ML workflows with clear promotion paths

Strategy 3: Version-based organization

Create projects for major model versions:

PCB-Detection-v1 (Initial release)
PCB-Detection-v2 (Improved architecture)
PCB-Detection-v3 (New dataset)

Best for: Long-term projects with significant version milestones

Strategy 4: Team-based organization

Organize by team member or responsibility:

PCB-Detection-Research
PCB-Detection-Engineering
PCB-Detection-QA

Best for: Large teams with specialized roles

Training project lifecycle

Follow this recommended lifecycle for managing training projects:

1. Creation phase

• Create training project with descriptive name
• Configure localization based on data residency requirements
• Document purpose in project description field
• Set up workflows for initial experiments

2. Development phase

• Run experiments with multiple workflow variations
• Monitor results and track metrics
• Iterate workflows based on performance
• Organize experiments with systematic naming

3. Production phase

• Identify best model from experiment results
• Download trained model for deployment
• Archive workflows by renaming with "PROD - " prefix
• Document configuration for team reference

4. Maintenance phase

• Clean up failed experiments by deleting unused workflows
• Keep successful experiments for reference
• Monitor deployed models for performance degradation
• Update workflows as needed for retraining

5. Completion phase

• Export final models before project deletion
• Document lessons learned for future projects
• Delete training project to clean up workspace
• Archive project documentation locally

Common project management scenarios

Resource management

Training projects consume Compute Credits based on GPU usage during training runs.

Monitoring resource usage

• Organization-wide — View total credit consumption in Resource Usage
• Per project — Track credits used by specific training projects
• Per run — See individual run costs in run history

Optimizing costs

Learn about resource usage and pricing →

Troubleshooting

Next steps

Related resources

• Create a Training Project — Set up new training workspaces
• Rename a Training Project — Update project names
• Delete a Training Project — Remove projects permanently
• Manage Workflows — Organize training configurations
• Manage Runs — Monitor training execution
• Evaluate a Model — Review training results
• Resource Usage — Monitor Compute Credits
• Team Settings — Manage organization access