Manage Models

Effective model management is essential for maintaining organized deployments and controlling access to your trained VLMs. Datature Vi provides comprehensive tools to rename, secure, and clean up models throughout your deployment lifecycle.

This guide covers all aspects of model management, from organizing model names and managing access keys to safely removing models you no longer need.

💡

Complete deployment workflow

Train model → Evaluate model → Download model → Manage models (you are here) → Run inference

Core model operations

Datature Vi provides three essential categories of model management operations:

Quick reference

Common model management tasks and where to find them:

Renaming models

Keep your deployments organized with descriptive, meaningful model names.

When to rename

• Improved clarity — Make model purpose clear to team members
• Version tracking — Distinguish between model versions (e.g., "PCB Defect v2.1")
• Deployment stages — Indicate status (e.g., "Production", "Staging", "Testing")
• Use case identification — Clearly describe what the model does

Key features

• Safe operation — Model ID remains unchanged; all references continue working
• No downtime — Active inference and deployments are unaffected
• Instant updates — Name changes appear immediately across the platform
• Unlimited changes — Rename as often as needed

Learn how to rename models →

Editing model keys

Manage unique identifiers assigned to your model checkpoints for better organization and reference.

What are model keys?

Model keys are unique identifiers assigned to model checkpoints. They help you organize and reference specific versions or checkpoints from your training runs.

When to edit keys

• Improved organization — Create more descriptive or meaningful identifiers
• Consistent naming — Align model keys with your team's naming conventions
• Version tracking — Use keys that reflect model versions or iterations
• Integration requirements — Match identifiers expected by your deployment systems

Key management features

• Flexible identifiers — Update checkpoint identifiers anytime
• Immediate updates — Changes appear instantly across the platform
• Unique references — Each checkpoint maintains its own distinct key
• Human-readable — Use meaningful names instead of auto-generated IDs

Learn how to edit model keys →

Deleting models

Permanently remove models you no longer need to keep your workspace organized and manage storage.

When to delete

• Superseded models — Remove older versions after deploying improved models
• Failed experiments — Clean up models from unsuccessful training runs
• Test models — Remove experimental or prototype models
• Storage optimization — Free up space by removing unused models
• Workspace cleanup — Maintain organized, clutter-free environment

Safety measures

Before deleting a model:

1. Download if needed — Download the model for backup or archival
2. Check deployments — Verify no active deployments depend on this model
3. Inform team — Notify collaborators about the planned deletion
4. Review carefully — Ensure you're deleting the correct model

🔒

Deletion is permanent and irreversible

• Model weights and adapters are deleted
• Model configuration is permanently removed
• Model ID becomes invalid
• Inference requests will fail
• Cannot be recovered through any means
• SDK queries for deleted model ID will fail

Always download a backup first if there's any chance you'll need the model later.

Learn how to delete models →

Best practices for model management

Model management workflow

Follow this recommended workflow for maintaining healthy model deployments:

1. Naming and identification

• Use consistent naming — Apply standard conventions (e.g., "TaskType-BaseModel-Version")
• Track versions — Clearly label model iterations
• Update identifiers — Keep model keys meaningful and descriptive
• Clear references — Use keys that indicate checkpoint purpose

2. Documentation

• Document purpose — Keep notes on what each model is for
• Separate environments — Distinguish production vs development models
• Version indicators — Include version info in model names and keys
• Maintain notes — Record changes and decisions

3. Maintenance

• Archive old models — Download before deleting
• Remove test models — Delete experimental models after evaluation
• Clean up failures — Remove models from unsuccessful training runs
• Monitor storage — Keep track of model storage usage

4. Deployment management

• Test before production — Validate models thoroughly before deployment
• Maintain backups — Keep downloaded copies of production models
• Version control — Track which model version is deployed where
• Document rollback plan — Keep previous versions accessible if needed

Programmatic model management

For advanced users, the Vi SDK provides programmatic access to model management operations:

import vi

client = vi.Client(
    secret_key="your-secret-key",
    organization_id="your-org-id"
)

# List all models
for run in client.runs:
    if run.status.phase == "succeeded":
        print(f"Model: {run.name} (Run ID: {run.run_id})")

# Download a model
downloaded = client.get_model(
    run_id="run-id",
    save_path="./models"
)

# Rename a model (via run)
run = client.runs.get(run_id="run-id")
run.update(name="New Model Name")

# Note: Model deletion is handled through the web interface
# for safety and confirmation requirements

Learn more about Vi SDK → | SDK Models API →

Accessing model management

All model management operations are accessed from the Models tab in your training project:

1. Navigate to your training project
2. Click the Models tab in the sidebar
3. Find the model you want to manage
4. Click the three-dot menu (⋮) next to the model
5. Select the desired action:
   • View Run — Review training details
   • Export Model — Download for deployment
   • Rename Model — Update model name
   • Edit Model Key — Update checkpoint identifier
   • Delete Model — Permanently remove model

Common questions

Troubleshooting

Next steps

Related resources

• Download a model — Export models for deployment and backup
• Train a model — Train VLMs on your datasets
• Evaluate a model — Assess model performance and accuracy
• Manage runs — View training history and metrics
• Manage workflows — Organize training configurations
• Run inference — Use models for predictions
• Vi SDK Getting Started — Programmatic model access