Duplicate a Workflow

Why duplicate workflows?

Duplicating workflows is essential for systematic experimentation and configuration management:

• A/B testing — Compare different configurations side-by-side
• Preserve working configurations — Keep a backup before making changes
• Template creation — Create base workflows for similar experiments
• Parameter experimentation — Test variations of successful configurations
• Team collaboration — Give team members starting points for their work

Common duplication scenarios:

How duplication works

When you duplicate a workflow, Datature Vi creates an exact copy with all configuration preserved:

What gets copied:

• System Prompt — Complete prompt instructions
• Dataset Configuration — Data source and split settings
• Model Settings — Architecture, parameters, and optimization settings
• Token Monitor values — Context window calculations
• Workflow canvas layout — Node positions and connections

What changes:

• Workflow name — " (Copy)" is appended to the original name
• Workflow ID — New unique identifier is generated
• Training history — Duplicate starts with no associated training runs
• Creation timestamp — Shows current date/time

💡

Independent workflows

The duplicate and original are completely independent. Changes to one don't affect the other, and they maintain separate training run histories.

Duplicate a workflow

Step 1: Navigate to your workflow

From your training project overview, locate the workflow you want to duplicate in the Workflows section.

Step 2: Open the workflow menu

Click the three-dot menu (⋮) on the right side of the workflow card to open the options menu.

The menu displays available actions:

• Duplicate — Create a copy of the workflow
• Delete — Permanently remove the workflow

Step 3: Select Duplicate

Click Duplicate from the menu.

The workflow is duplicated immediately without additional confirmation.

✅

Workflow duplicated successfully

A new workflow appears in your Workflows section with " (Copy)" appended to the name. You can now modify this copy independently.

Step 4: Rename the duplicate (recommended)

The duplicated workflow is created with " (Copy)" appended to the original name:

• Original: PCB-Component-Detection
• Duplicate: PCB-Component-Detection (Copy)

Best practice: Rename the duplicate immediately to reflect its purpose:

• ✅ PCB-Component-Detection-Detailed-Prompt-v2
• ✅ PCB-Component-Detection-NVLM-8B
• ✅ PCB-Component-Detection-LR0.0001
• ❌ PCB-Component-Detection (Copy)
• ❌ PCB-Component-Detection (Copy) (Copy)

Step 5: Modify the duplicate

Click the duplicated workflow to open the workflow canvas, then modify any component:

• System Prompt — Test different instruction styles or detail levels
• Dataset — Change data splits or select different datasets
• Model — Switch architectures or adjust training parameters

Save your changes when finished.

Step 6: Run training (optional)

Start a training run using the duplicated workflow to compare results with the original:

1. Click Run Training from the workflow canvas, or
2. Navigate to Runs and select the duplicated workflow

Learn how to start training runs →

Duplication strategies for experimentation

Use these systematic approaches to experiment with workflow variations:

Strategy 1: System prompt A/B testing

Create multiple duplicates with different prompt variations:

Baseline workflow:

Product-Detection-Baseline

Duplicated variations:

Product-Detection-Brief-Prompt
Product-Detection-Detailed-Prompt
Product-Detection-Technical-Prompt
Product-Detection-Zero-Shot-Prompt

How to execute:

1. Duplicate baseline workflow 4 times
2. Rename each duplicate with prompt identifier
3. Edit each duplicate with different system prompt
4. Run all workflows simultaneously
5. Compare results to identify best prompt

Strategy 2: Model architecture comparison

Test different model architectures with identical prompts and data:

Baseline workflow:

Defect-Detection-Baseline

Duplicated variations:

Defect-Detection-Qwen2.5-7B
Defect-Detection-NVLM-8B
Defect-Detection-InternVL2.5-4B
Defect-Detection-Molmo-7B

How to execute:

1. Duplicate baseline workflow for each architecture
2. Rename each duplicate with model name
3. Edit model selection in each duplicate
4. Keep system prompt and dataset identical
5. Compare performance metrics across architectures

Strategy 3: Hyperparameter sweeps

Systematically test learning rates, batch sizes, or other parameters:

Baseline workflow:

Training-Baseline

Duplicated variations:

Training-LR0.00001
Training-LR0.0001
Training-LR0.001
Training-LR0.01

How to execute:

1. Duplicate baseline for each parameter value
2. Rename each duplicate with parameter value
3. Edit training settings in each duplicate
4. Run systematic parameter sweep
5. Identify optimal hyperparameters

Strategy 4: Dataset experiment tracking

Test same model with different data configurations:

Baseline workflow:

PCB-Model-Baseline

Duplicated variations:

PCB-Model-Dataset-v1
PCB-Model-Dataset-v2-Augmented
PCB-Model-Dataset-v3-Filtered
PCB-Model-Dataset-Combined

How to execute:

1. Duplicate baseline for each dataset version
2. Rename each duplicate with dataset identifier
3. Edit dataset selection in each duplicate
4. Keep model and prompt consistent
5. Evaluate impact of dataset changes

When to duplicate vs. edit

Understanding when to duplicate versus editing the original workflow:

Duplicate the workflow when:

✅ Testing variations — Comparing different configurations side-by-side ✅ A/B testing — Running systematic experiments ✅ Preserving working config — Keeping a backup before major changes ✅ Different experiments — Trying significantly different approaches ✅ Team experimentation — Giving team members their own versions ✅ Historical reference — Maintaining configuration snapshots

Edit the existing workflow when:

✅ Fixing errors — Correcting mistakes in configuration ✅ Minor adjustments — Small parameter tweaks ✅ Updating dataset — Pointing to refreshed data ✅ No comparison needed — Not running side-by-side tests ✅ Pre-training phase — Workflow hasn't been used for training yet ✅ Iterative refinement — Continuous improvement without branching

📘

General guideline

If you want to compare results, duplicate. If you just want to improve the existing workflow, edit.

Duplicate vs. rename vs. create new

Understand when to use each workflow management operation:

Example scenarios:

Managing duplicated workflows

Naming duplicates effectively

Use descriptive names that indicate what's different:

Format: {Base-Name}-{Variation}-{Version}

Examples:

Learn more about workflow naming →

Organizing experiment series

Keep related duplicates organized with consistent naming:

Method 1: Version numbering

Product-Recognition-v1
Product-Recognition-v2
Product-Recognition-v3
Product-Recognition-v4

Method 2: Variation identifiers

Defect-Detection-Brief
Defect-Detection-Standard
Defect-Detection-Detailed
Defect-Detection-Technical

Method 3: Parameter values

Training-Batch16-Epoch5
Training-Batch32-Epoch5
Training-Batch16-Epoch10
Training-Batch32-Epoch10

Cleaning up after experiments

After completing your experiments:

1. Identify the winner — Determine which duplicate performed best
2. Rename the winner — Update name to indicate it's the chosen configuration (e.g., "PROD - Model Name")
3. Archive alternatives — Rename unsuccessful duplicates with "ARCHIVED - " prefix
4. Delete obvious failures — Remove configurations that clearly didn't work
5. Keep interesting variations — Retain configurations that might be useful for future reference

Duplicating workflows via workflow canvas

You can also duplicate workflows when viewing the workflow canvas:

From the workflow canvas:

1. Open the workflow you want to duplicate
2. Click the workflow menu in the top-right corner
3. Select Duplicate Workflow
4. The duplicate is created and you're redirected to the new workflow canvas

This method is useful when you're already working within a workflow and decide you want to create a variation.

Common questions

Best practices for workflow duplication

Duplication workflows for common tasks

Task 1: Backup before major changes

Scenario: You have a working workflow but want to make significant changes.

Process:

1. Duplicate the working workflow
2. Rename duplicate to [Original-Name]-Backup
3. Make changes to the original workflow
4. Test the modified workflow
5. If changes work: Delete the backup
6. If changes fail: Revert to backup or continue working

Task 2: Progressive iteration

Scenario: Improving a workflow through multiple iterations.

Process:

1. Start with baseline: Model-v1
2. Train and evaluate Model-v1
3. Duplicate Model-v1 → rename to Model-v2
4. Make improvements to Model-v2
5. Train and compare v1 vs v2
6. Continue: Duplicate Model-v2 → Model-v3
7. Keep successful versions, archive or delete unsuccessful ones

Task 3: Team experimentation

Scenario: Multiple team members testing different approaches.

Process:

1. Create baseline workflow: PCB-Detection-Baseline
2. Each team member duplicates the baseline
3. Rename with owner prefix: Wei-PCB-Prompt-Test, Sarah-PCB-Model-Test
4. Each member experiments independently
5. Team compares results in meeting
6. Choose best approach and create production workflow

Troubleshooting

Next steps

Related resources

• Create a Workflow — Set up new training workflows from scratch
• Rename a Workflow — Update workflow names for better organization
• Delete a Workflow — Remove workflows you no longer need
• Manage Workflows — Overview of all workflow management operations
• Manage Runs — Start and monitor training runs