Download this tutorial as a Jupyter notebook

Safe Synthesizer 101#

Learn the fundamentals of NeMo Safe Synthesizer by creating your first Safe Synthesizer job using provided defaults. In this tutorial, you’ll upload sample customer data, replace personally identifiable information, fine-tune a model, generate synthetic records, and review the evaluation report.

Prerequisites#

Before you begin, make sure that you have:

  • Access to a deployment of NeMo Safe Synthesizer (see Getting Started with NeMo Safe Synthesizer)

  • Python environment with nemo-platform SDK installed

  • Basic understanding of Python and pandas

  • An NVIDIA GPU on the host machine (80GB+ VRAM recommended — check with nvidia-smi)

What You’ll Learn#

By the end of this tutorial, you’ll understand how to:

  • Upload datasets for processing

  • Run Safe Synthesizer jobs using the Python SDK

  • Track job progress and retrieve results

  • Interpret evaluation reports

Step 1: Install the SDK#

Install the NeMo Platform SDK with Safe Synthesizer support. Run the following command in a terminal (shell):

if command -v uv &> /dev/null; then
    uv pip install nemo-platform[safe-synthesizer] kagglehub matplotlib
else
    pip install nemo-platform[safe-synthesizer] kagglehub matplotlib
fi

Step 2: Configure the Client#

Set up the client to connect to your Safe Synthesizer deployment:

import os
from nemo_platform import NeMoPlatform

# Configure the client
client = NeMoPlatform(
    base_url=os.environ.get("NMP_BASE_URL", "http://localhost:8080"),
    workspace="default",
    access_token=os.environ.get("NMP_ACCESS_TOKEN"),
)
# set to none by default, update it if you need an hf_token
hf_secret_name = None

print("✅ Client configured successfully")

Step 3: Verify Service Connection#

Test the connection to ensure Safe Synthesizer is accessible:

try:
    jobs = client.safe_synthesizer.jobs.list(workspace="default")
    print("✅ Successfully connected to Safe Synthesizer service")
    print(f"Found {len(jobs.data)} existing jobs")
except Exception as e:
    print(f"❌ Cannot connect to service: {e}")
    print("Please verify base_url and service status")

Step 4: Load Sample Dataset#

For this tutorial, we’ll use a women’s clothing reviews dataset from Kaggle that contains some PII:

import pandas as pd
import kagglehub

# Download the dataset
path = kagglehub.dataset_download("nicapotato/womens-ecommerce-clothing-reviews")
df = pd.read_csv(f"{path}/Womens Clothing E-Commerce Reviews.csv", index_col=0)

print(f"✅ Loaded dataset with {len(df)} records")
print("\nDataset preview:")
print(df.head())

Dataset details:

  • Contains customer reviews of women’s clothing

  • Includes age, product category, rating, and review text

  • Some reviews contain PII like height, weight, age, and location

Step 5: Configure Column Classification#

Before running jobs, set up column classification for accurate PII detection.

Tip

Column classification uses an LLM to automatically detect column types and improve PII detection accuracy. Without this setup, you may see classification errors and reduced detection quality.

Note

The platform pre-configures a system/nvidia-build model provider during startup. This provider routes inference requests to models hosted on build.nvidia.com using the API base URL https://integrate.api.nvidia.com and the NGC API key with Public API Endpoints permissions provided during deployment (automatically saved as the built-in system/ngc-api-key secret).

You can verify this provider exists by running nmp inference providers list --workspace system.

The tutorials in these docs use this provider for inference, but you can alternatively create your own and use it instead.

# Use the pre-configured NVIDIA Build model provider
# This provider is set up automatically during platform deployment
provider_name = "system/nvidia-build"
print(f"✅ Using model provider: {provider_name}")

Note

If you prefer not to send column data to build.nvidia.com, you can deploy your own LLM and create a custom model provider. Pass the fully-qualified provider name (workspace/provider-name) to .with_classify_model_provider() instead.

Step 6: HuggingFace Token Usage (Optional)#

If you’re using private HuggingFace models or want to avoid rate limits, create a secret for your HuggingFace token:

import os
import time

# Create a unique secret name (use hyphens, not underscores)
hf_secret_name = f"hf-token-{int(time.time())}"
hf_token = os.environ.get("HF_TOKEN")

if hf_token:
    # Store your HuggingFace token as a platform secret
    client.secrets.create(
        workspace="default",
        name=hf_secret_name,
        data=hf_token
    )
    print(f"✓ Created secret: {hf_secret_name}")

Step 7: Create and Run a Safe Synthesizer Job#

Use the SafeSynthesizerJobBuilder to configure and create a job:

import pandas as pd
from nemo_platform.beta.safe_synthesizer.job_builder import SafeSynthesizerJobBuilder

# Create a project for our jobs (creates if it doesn't exist)
project_name = "test-project"
try:
    client.projects.create(workspace="default", name=project_name)
except Exception:
    pass  # Project may already exist

# Build the job configuration
job_name = f"synthesis-test-{pd.Timestamp.now().strftime('%Y%m%d-%H%M%S')}"
builder = (
    SafeSynthesizerJobBuilder(client)
    .with_data_source(df)
    .with_classify_model_provider(provider_name)  # Enable column classification
    .with_replace_pii()  # Enable PII replacement
    .synthesize()  # Enable synthesis
)

if hf_secret_name:
    # add the token secret if an HF token was specified
    builder = builder.with_hf_token_secret(hf_secret_name)

# Create and start the job
job = builder.create_job(name=job_name, project=project_name)
print(f"✅ Job created: {job.job_name}")

What happens next:

  1. Dataset is uploaded to the fileset storage

  2. PII detection and replacement

  3. Model fine-tuning on your data

  4. Synthetic data generation

  5. Quality and privacy evaluation

Step 8: Monitor Job Progress#

Check the job status:

status = job.fetch_status()
print(f"Current status: {status}")

Job States:

  • created: Job has been created

  • pending: Waiting for GPU resources

  • active: Processing your data

  • completed: Finished successfully

  • error: Encountered an error

View real-time logs:

job.print_logs()

Wait for completion (this may take 15-30 minutes depending on data size):

print("⏳ Waiting for job to complete...")
try:
    job.wait_for_completion()
    print("✅ Job completed!")
except RuntimeError as e:
    print(f"❌ Job failed: {e}")
    raise

Note

wait_for_completion() raises RuntimeError if the job ends in an error or cancelled state. Check the printed status output and logs above for the cause.

Step 9: Retrieve Synthetic Data#

Once the job is complete, retrieve the generated synthetic data:

synthetic_df = job.fetch_data()

print(f"✅ Generated {len(synthetic_df)} synthetic records")
print("\nSynthetic data preview:")
print(synthetic_df.head())

Compare with original data structure:

print("\n📊 Data Comparison:")
print(f"Original shape: {df.shape}")
print(f"Synthetic shape: {synthetic_df.shape}")
print(f"\nOriginal columns: {list(df.columns)}")
print(f"Synthetic columns: {list(synthetic_df.columns)}")

Step 10: Review Evaluation Report#

Fetch the job summary with high-level metrics:

summary = job.fetch_summary()

print("📈 Evaluation Summary:")
print(f"  Synthetic Quality Score: {summary.synthetic_data_quality_score}")
print(f"  Data Privacy Score: {summary.data_privacy_score}")
print(f"  Valid Records: {summary.num_valid_records}/{summary.num_prompts}")

Download the full HTML evaluation report:

job.save_report("./evaluation_report.html")
print("✅ Evaluation report saved to evaluation_report.html")

If using Jupyter, display the report inline:

job.display_report_in_notebook()

The evaluation report includes:

  • Synthetic Quality Score (SQS): Measures data utility

    • Column correlation stability

    • Deep structure stability

    • Column distribution stability

    • Text semantic similarity

    • Text structure similarity

  • Data Privacy Score (DPS): Measures privacy protection

    • Membership inference protection

    • Attribute inference protection

    • PII replay detection

Understanding the Results#

Interpreting Scores#

The evaluation report contains two high-level scores: Synthetic Quality Score (SQS) and Data Privacy Score (DPS). Both are measured out of 10, and higher is better. To learn more about how to interpret the scores, refer to the evaluation guide.

Next Steps#

Now that you’ve completed your first Safe Synthesizer job, explore more advanced features:

Advanced Tutorials#

Documentation#

Try These Next#

  1. Customize PII replacement: Configure specific entity types and replacement strategies

  2. Enable differential privacy: Add formal privacy guarantees with epsilon and delta parameters

  3. Tune generation parameters: Adjust temperature and sampling for better synthetic data

  4. Use your own data: Replace the sample dataset with your sensitive data

Cleanup#

List and optionally delete completed jobs:

# List all jobs
all_jobs = client.safe_synthesizer.jobs.list(workspace="default")
print(f"Total jobs: {len(all_jobs.data)}")

# Delete this job (optional)
# client.safe_synthesizer.jobs.delete(job.job_name, workspace="default")
# print(f"✅ Job {job.job_name} deleted")

Troubleshooting#

Common Issues#

Connection errors:

  • Verify NMP_BASE_URL is correct

  • Check that Safe Synthesizer service is running

  • Ensure network connectivity

Job failures:

  • Check logs with job.print_logs()

  • Verify dataset format (CSV with proper columns)

  • Ensure sufficient GPU memory for model size

Slow performance:

  • Reduce dataset size for testing

  • Use smaller model (adjust training.pretrained_model)

  • Check GPU availability

For more help, see Safe Synthesizer Jobs.

Error: “Dataset must have at least 200 records to use holdout.”

This occurs when synthesis is enabled on datasets with fewer than 200 records. Holdout validation splits your data into training and test sets to measure quality, requiring a minimum dataset size.

Solution:

builder = (
    SafeSynthesizerJobBuilder(client)
    .with_data_source(df)
    .with_data(holdout=0)  # Disable holdout for small datasets
    .with_replace_pii()
    .synthesize()
)

Warning

Disabling holdout means you won’t get quality metrics like privacy scores and synthetic data quality scores. For production use, ensure your dataset has at least 200 records.