AI Frameworks · hitsz-ids

synthetic-data-generator

SDG is a Python framework for generating high-quality synthetic tabular data using statistical models (CTGAN, GaussianCopula) and LLM-based approaches. It supports privacy-enhancing techniques and is optimized for large datasets, making it suitable for data sharing, testing, and model training without exposing sensitive information.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/hitsz-ids/synthetic-data-generator

2.4k

GitHub stars

391

Forks

Python

Primary language

Apache-2.0

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	hitsz-ids/synthetic-data-generator
Owner	hitsz-ids
Primary language	Python
License	Apache-2.0 — OSI-approved
Stars	2.4k
Forks	391
Open issues	27
Latest release	0.2.4 (2024-12-03)
Last updated	2026-05-25
Source	https://github.com/hitsz-ids/synthetic-data-generator

What synthetic-data-generator is

SDG provides pluggable synthesis models (GAN-based, statistical, LLM-based), a data processor pipeline for format conversion and null handling, metadata management for single and multi-table schemas, and differential privacy support. Built in Python with memory-optimized implementations targeting billion-row datasets.

Quickstart

Get the synthetic-data-generator source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/hitsz-ids/synthetic-data-generator.git2cd synthetic-data-generator3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Privacy-Compliant Data Sharing

Generate synthetic datasets that retain statistical properties of production data without sensitive information, enabling safe sharing for analysis, reporting, or third-party collaboration under GDPR/CCPA constraints.

Testing & Development Data

Create realistic synthetic datasets for integration testing, system prototyping, and QA environments without duplicating or exposing real customer/patient data.

Model Training & Augmentation

Supplement training datasets or generate entirely synthetic cohorts to address class imbalance, data scarcity, or privacy concerns in machine learning workflows.

Implementation considerations

Model selection (CTGAN vs. GaussianCopula vs. LLM) depends on data shape, size, cardinality, and privacy budget; no single model is optimal for all scenarios.
Data processor pipeline requires configuration for datetime, categorical, and null handling; automatic inference is available but should be validated against domain logic.
LLM-based synthesis requires external API keys (e.g., OpenAI) and incurs latency and cost per generation; offline statistical models are more cost-effective for high-volume synthetic data.
Memory footprint for large datasets is significantly reduced vs. alternatives (per benchmarks), but CTGAN with high-cardinality categorical columns still requires testing on target hardware.
Privacy guarantees depend on model choice and configuration; differential privacy is supported but requires careful hyperparameter tuning to balance privacy and utility.

When to avoid it — and what to weigh

Unstructured Data Focus — SDG is specialized for tabular data; it is not designed for text, images, audio, or other unstructured modalities.
Real-Time Synthesis Requirements — Model training and inference are batch-oriented; if you need sub-second synthetic data generation in production, this framework is not optimized for that use case.
Complex Multi-Table Relationships — While metadata supports multi-table description, the framework's primary focus is single-table synthesis; complex foreign key dependencies and referential integrity constraints require careful custom implementation.
Zero Trust in Open Source Supply Chain — If your organization requires extensive security audits and cannot accept risks from community-maintained dependencies (e.g., GAN libraries, LLM integrations), conduct full assessment before adoption.

License & commercial use

Apache License 2.0 (Apache-2.0). Permissive OSI-approved license allowing commercial use, modification, and distribution with proper attribution and liability disclaimer.

Apache-2.0 permits commercial use, but verify that: (1) LLM integrations (e.g., GPT models) do not violate OpenAI or third-party ToS; (2) your organization's use of synthetic data from real customer data complies with data protection laws; (3) you retain appropriate liability insurance and legal review for sensitive use cases (healthcare, financial).

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Good
Assessment confidence	Medium

Security considerations

SDG is designed to output privacy-preserving synthetic data, not to encrypt or secure production data in transit. Key considerations: (1) input data handling—raw production data is processed during synthesis; ensure secure data pipelines and access controls; (2) differential privacy support is opt-in and requires careful tuning; default synthesis may still leak statistical properties of training data; (3) LLM integrations may transmit data summaries to external APIs; review API privacy policies; (4) no built-in audit logging or data lineage tracking; implement external logging if required for compliance; (5) dependency supply chain risk—CTGAN, PyTorch, scikit-learn are external; pin versions and scan for CVEs.

Alternatives to consider

Mostly AI

Commercial closed-source platform with strong UI/UX, automatic model selection, and enterprise SLAs; better for non-technical users but higher cost and lock-in risk.

Synthesized.io (Synthesized SDK)

Enterprise-grade synthetic data platform with stronger documentation, multi-table support, and differential privacy guarantees; higher price point but more mature governance features.

SDV (Synthetic Data Vault) by DataCebo

Apache-2.0 open-source alternative with broader model ecosystem and multi-table support; SDG claims lower memory usage for CTGAN on large datasets, but SDV may be more mature for general-purpose use.

Software development agency

Build on synthetic-data-generator with DEV.co software developers

Review the GitHub repository, run the colab examples, and benchmark memory/performance on your dataset size before committing to production. Consult legal/privacy teams for regulatory compliance validation.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

mostlyai DB-GPT CAG

Related on DEV.co

Explore the category and the services that help you build with it.

AI Frameworks AI Application Development API Development Custom Software Development Talk to DEV.co

synthetic-data-generator FAQ

Does SDG guarantee that synthetic data is not re-identifiable?

No. SDG generates synthetic data that statistically resembles the training data. Without differential privacy enabled and tuned, reconstruction attacks or re-identification may be possible. Differential privacy is optional and must be explicitly configured.

Can I use SDG to synthesize data without any real training data (zero-shot)?

Yes, via the LLM-based SingleTableGPTModel. You provide only metadata (column names, types, ranges) and the LLM generates plausible synthetic data. Utility and coverage depend on LLM knowledge and metadata quality.

What is the typical memory overhead for training CTGAN on a 1 million row dataset?

Unknown from provided data. README mentions billion-level data support and lower memory vs. SDV, with benchmarks in the repo. Contact maintainers or run local benchmarks on your hardware for precise estimates.

Does SDG support generating data with temporal or sequential patterns?

Not explicitly mentioned in the README for time-series synthesis. Data processor supports datetime column conversion, but whether CTGAN preserves temporal correlations is not stated. Requires documentation review or testing.

Software development & web development with DEV.co

Need help beyond evaluating synthetic-data-generator? DEV.co is a software development agency offering software development services and web development for teams of every size. Our software developers and web developers build custom software, web applications, APIs, and ai frameworks integrations — and maintain them long-term.

Evaluate SDG for Your Data Synthesis Needs

Talk to DEV.co Schedule a Consultation