Manage Workflows

Effective workflow management keeps your training projects organized and helps you track experiments systematically. Datature Vi provides tools to rename, duplicate, edit, and delete workflows as your project evolves.

This guide covers all aspects of workflow management, from organizing configurations with descriptive names to cleaning up unused workflows and creating variations for experimentation.

💡

Complete training workflow

Create workflow → Manage workflows (you are here) → Start training run → Evaluate model

What are workflows?

Workflows are reusable training configurations that define how your VLM learns from your data. Each workflow specifies:

• System Prompt — Instructions that guide your VLM's behavior
• Dataset Configuration — Data source and splitting strategy
• Model Settings — Architecture, training parameters, and optimization settings

Once created, workflows can be:

• Reused for multiple training runs with consistent settings
• Edited to refine configurations based on results
• Duplicated to create variations for A/B testing
• Renamed for better organization and clarity
• Deleted when no longer needed

📘

Workflows vs. Training Runs

• Workflow — A saved configuration template (what you plan to train)
• Training Run — An actual training session using a workflow (the training execution)

Think of workflows as recipes and training runs as the meals you cook from those recipes.

Core workflow operations

Datature Vi provides four essential workflow management operations:

Quick reference

Common workflow management tasks and where to find them:

Editing workflows

Modify existing workflow configurations to refine your training settings based on results and experiments.

When to edit

• After initial runs — Adjust parameters based on training performance
• Fix configuration errors — Correct mistakes in system prompts or settings
• Update datasets — Switch to updated or expanded datasets
• Optimize parameters — Fine-tune learning rates, batch sizes, or epochs

Key features

• Safe operation — Editing a workflow doesn't affect existing training runs
• Version snapshots — Each run captures the workflow configuration at that time
• No downtime — Make changes without interrupting active runs
• Flexible updates — Modify any component (prompt, dataset, or model settings)

📘

Past runs remain unchanged

Training runs capture a snapshot of the workflow configuration when started. Editing the workflow afterwards doesn't affect completed or in-progress runs.

How to edit

1. Navigate to your training project
2. Open the workflow from the Workflows section
3. Make changes to any component (System Prompt, Dataset, or Model)
4. Save the updated workflow

Learn how to edit workflows →

Renaming workflows

Keep your training project organized with clear, descriptive workflow names that reflect their purpose.

When to rename

• Improved clarity — Make workflow purpose clear to team members
• Naming conventions — Apply systematic naming standards
• Experiment tracking — Include version numbers or test parameters
• From "Untitled" — Rename default workflows before first training run

Naming strategies

Different naming conventions for different needs:

Key features

• Instant updates — Name changes appear immediately across platform
• Safe operation — Workflow ID and connections remain unchanged
• No limits — Rename as often as needed
• Unlimited flexibility — Use any naming convention that works for your team

Learn how to rename workflows →

Duplicating workflows

Create copies of workflows to experiment with variations while preserving the original configuration.

When to duplicate

• A/B testing — Compare different system prompts or model architectures
• Parameter experimentation — Test learning rates, batch sizes, or epochs
• Backup before editing — Preserve working configurations
• Template creation — Create base workflows for similar projects
• Team sharing — Give team members starting points for their experiments

Common duplication scenarios

How to duplicate

1. Navigate to your training project
2. Locate the workflow to duplicate
3. Click the three-dot menu (⋮) on the workflow card
4. Select Duplicate
5. A copy is created with " (Copy)" appended to the name
6. Rename and modify the duplicate as needed

Learn how to duplicate workflows →

Deleting workflows

Permanently remove workflows you no longer need to keep your project organized and focused on active configurations.

When to delete

• Failed experiments — Configurations that didn't produce useful results
• Duplicate workflows — Accidentally created copies
• Outdated configurations — Replaced by improved versions
• Project cleanup — Remove obsolete workflows after project completion

Safety measures

Before deleting a workflow:

⚠️

Consider duplicating first

If the workflow might be useful later:

• Duplicate the workflow as a backup
• Rename for archival — Add "ARCHIVED - " prefix instead of deleting
• Keep for reference — Workflows don't consume resources; keeping them provides training history context

What gets deleted

Deleted immediately:

• Workflow configuration (system prompt, dataset settings, model parameters)
• Workflow name, description, and metadata
• Workflow canvas layout

Not affected:

• Completed training runs — Remain viewable with results and metrics
• Trained models — Continue functioning normally
• Source dataset — Remains intact and usable
• Other workflows — Unaffected by deletion

💡

Training runs remain accessible

Deleting a workflow doesn't delete training runs that used it. Your training history, metrics, and models remain fully accessible.

Cannot delete workflows with active runs

The platform prevents deletion of workflows currently being used by running training jobs. You must:

1. Wait for the run to complete, or
2. Cancel the training run first
3. Then delete the workflow

This safeguard prevents corrupting active training runs.

Learn how to delete workflows →

Best practices for workflow management

Workflow organization strategies

Organize your workflows systematically to track experiments and maintain clarity as your project grows.

Strategy 1: Iterative versioning

Keep a linear progression of improvements:

PCB-Detection-v1 (Baseline)
PCB-Detection-v2 (Improved prompt)
PCB-Detection-v3 (Different model)
PCB-Detection-v4 (Optimized parameters)

Best for: Single-track development with incremental improvements

Strategy 2: Parallel experimentation

Test multiple variations simultaneously:

PCB-Detection-Qwen2.5-7B-LoRA
PCB-Detection-NVLM-8B-LoRA
PCB-Detection-InternVL2.5-4B-LoRA

Best for: Architecture comparison and model selection

Strategy 3: Parameter sweeps

Systematically test hyperparameters:

PCB-Detection-LR0.0001
PCB-Detection-LR0.0005
PCB-Detection-LR0.001
PCB-Detection-LR0.005

Best for: Hyperparameter optimization and tuning

Strategy 4: Task-based organization

Organize by different tasks or use cases:

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

Best for: Multi-task projects with distinct objectives

Workflow lifecycle

Follow this recommended workflow management lifecycle:

1. Creation phase

• Create workflow with default settings
• Rename immediately from "Untitled Workflow" to descriptive name
• Document purpose in workflow description field
• Start initial run to establish baseline

2. Iteration phase

• Review training results from initial run
• Duplicate workflow for experimentation
• Edit copy with refined settings
• Run comparison to evaluate improvements
• Update naming to reflect versions (v1, v2, v3)

3. Production phase

• Identify best performer from experiments
• Rename clearly to indicate production status (e.g., "PROD - PCB Detection")
• Archive alternatives by renaming with "ARCHIVED - " prefix
• Document configuration for team reference

4. Maintenance phase

• Delete failed experiments that won't be referenced
• Keep archived workflows for historical context
• Update production workflow as needed based on ongoing results
• Create new experiments when requirements change

Common workflow management scenarios

Troubleshooting

Next steps

Related resources

• Create a Workflow — Set up new training configurations
• Rename a Workflow — Update workflow names for better organization
• Duplicate a Workflow — Create workflow copies for experimentation
• Delete a Workflow — Remove workflows you no longer need
• Manage Runs — Monitor and control training execution
• Evaluate a Model — Review training results
• Configure Training Settings — Advanced parameter tuning
• Resource Usage — Monitor compute and storage consumption