# 🛡️ NeMo Safe Synthesizer
NVIDIA NeMo Safe Synthesizer creates private, safe versions of sensitive tabular datasets -- entirely synthetic data with no one-to-one mapping to your original records. Purpose-built for privacy compliance and sensitive information protection while preserving data utility for downstream AI tasks.
## Quick Start
<a href="https://brev.nvidia.com/launchable/deploy/now?launchableID=env-3HBtA2NKQaBukL2TyDphWUcvQ17">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://brev-assets.s3.us-west-1.amazonaws.com/nv-lb-light.svg">
<img alt="Launch on Brev" src="https://brev-assets.s3.us-west-1.amazonaws.com/nv-lb-dark.svg">
</picture>
</a>
Try it without installing anything: the launchable above deploys a GPU instance with NeMo Safe Synthesizer and the tutorial notebooks preinstalled. Note that the instance bills continuously and cannot be paused -- delete it when you are finished.
Read detailed usage below, or jump to the documentation with [Getting Started](https://nvidia-nemo.github.io/Safe-Synthesizer/user-guide/getting-started/) or the [Safe Synthesizer 101](https://nvidia-nemo.github.io/Safe-Synthesizer/tutorials/safe-synthesizer-101/) notebook.
### Prerequisites
- Python 3.11–3.14 (`.python-version` pins 3.13 for local/dev bootstrap; any 3.11, 3.12, 3.13, or 3.14 interpreter works)
- [uv](https://docs.astral.sh/uv/) -- Python package manager
- NVIDIA GPU (A100 or larger) for training and generation
- Linux only -- macOS, Windows, and Apple Silicon are not supported for training or generation. A CPU-only install is available for development and configuration validation.
### Installation
Download the installer from the latest stable GitHub
release. The released script pins NeMo Safe Synthesizer, its constraints, and
the package indexes that support that version.
```bash
curl -fsSLO https://github.com/NVIDIA-NeMo/Safe-Synthesizer/releases/latest/download/install_nss.sh
chmod +x install_nss.sh
UV_PROJECT_ENVIRONMENT=.venv ./install_nss.sh
source .venv/bin/activate
```
> [!NOTE]
> This project will download and install additional third-party open source
> software projects. Review the license terms of these open source projects
> before use.
The installer uses CUDA 12.9 by default. To choose another runtime, set
`CUDA`:
```bash
CUDA=130 ./install_nss.sh # CUDA 13.0
CUDA=cpu ./install_nss.sh # CPU-only development and validation
```
CUDA 13.0 requires an NVIDIA driver version 580.65.06 or newer.
To customize the package source or installation command, use `DRY_RUN=1` with the desired
`CUDA` setting. The installer prints the complete `uv pip` command without
running it. Copy that command, edit it, and run it manually. For example,
```bash
DRY_RUN=1 CUDA=130 ./install_nss.sh
```
Or install from source:
```bash
git clone https://github.com/NVIDIA-NeMo/Safe-Synthesizer.git
cd Safe-Synthesizer
make setup # installs pinned mise, pinned tools from mise.lock, and .venv
mise run bootstrap-nss cuda
```
### Run the public container
The published GPU runtime is available from [GitHub Container Registry](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/pkgs/container/safe-synthesizer) at `ghcr.io/nvidia-nemo/safe-synthesizer`.
It runs the CLI without a local Python installation; see the [Docker guide](https://nvidia-nemo.github.io/Safe-Synthesizer/user-guide/docker/) for image tags, mounts, GPU access, and configuration.
Development tools (`dprint`, `ruff`, `ty`, `yq`, `gh`, etc.) are managed via [mise](https://mise.jdx.dev/). Tool versions are declared in `.mise.toml` and locked in `mise.lock` (committed). mise also manages environment variables -- place project-local secrets or overrides in `.env` or `.env.local` (both git-ignored, auto-loaded by mise).
For IDEs to discover mise-managed tools, [add mise shims to the `PATH` in your default shell profile](https://mise.jdx.dev/ide-integration.html#adding-shims-to-path-in-your-default-shell).
Project commands run through mise tasks under `.mise/tasks/`: `*.toml` files for declarative tasks, executable scripts for bash-heavy logic.
```bash
mise tasks # list public tasks
mise tasks --hidden # include helper and legacy alias tasks
mise tasks deps check # inspect the read-only quality-check graph
mise run check ::: test # local pre-PR gate: static checks + unit tests
```
### Running
Activate Python virtual environment and run the CLI using `safe-synthesizer`:
```bash
> safe-synthesizer --help
Usage: safe-synthesizer [OPTIONS] COMMAND [ARGS]...
NeMo Safe Synthesizer command-line interface. This application is used to
run the Safe Synthesizer pipeline. It can be used to train a model, generate
synthetic data, and evaluate the synthetic data. It can also be used to
modify a config file.
Options:
--help Show this message and exit.
Commands:
artifacts Artifacts management commands.
config Manage Safe Synthesizer configurations.
run Run the Safe Synthesizer end-to-end pipeline.
```
## Running the Pipeline
The `run` command executes the Safe Synthesizer pipeline. Without a subcommand, it runs the full end-to-end pipeline:
```bash
> uv run safe-synthesizer run --help
Usage: safe-synthesizer run [OPTIONS] COMMAND [ARGS]...
Run the Safe Synthesizer end-to-end pipeline.
Without a subcommand, runs the full end-to-end pipeline. Use 'run train' or
'run generate' for individual stages.
Options:
--config TEXT path to a yaml config file
--data-source TEXT Dataset name, URL, or path to CSV dataset.
For 'run generate', this is optional if a
cached dataset exists in the workdir.
--artifact-path DIRECTORY Base directory for all runs. Runs are
created as <artifact-
path>/<config>---<dataset>/<timestamp>/. Can
also be set via NSS_ARTIFACTS_PATH env var.
[default: ./safe-synthesizer-artifacts]
--run-path DIRECTORY Explicit path for this run's output
directory. When specified, outputs go
directly to this path. Overrides --artifact-
path.
--output-file PATH Path to output CSV file. Overrides the
default workdir output location.
--log-format [json|plain] Log format for console output. File logging
will always be JSON. Can also be set via
NSS_LOG_FORMAT env var. [default: plain]
--log-color / --no-log-color Whether to colorize the log output on the
console. [default: --log-color]
--log-file PATH Path to log file. Defaults to a file nested
under the run directory. Can also be set via
NSS_LOG_FILE env var.
--wandb-mode [online|offline|disabled]
Wandb mode. 'online' will upload logs to
wandb, 'offline' will save logs to a local
file, 'disabled' will not upload logs to
wandb. Can also be set via WANDB_MODE env
var. [default: disabled]
--wandb-project TEXT Wandb project. Can also be set via
WANDB_PROJECT env var.
-v Verbose logging. 'v' shows debug info from
main program, 'vv' shows debug from
dependencies too
--dataset-registry TEXT URL or path of a dataset registry YAML file.
If provided, datasets in the registry may be
referenced by name in --data-source. Can also be set
via NSS_DATASET_REGISTRY env var. If both
env var and CLI option are provided, the CLI
option takes precedence.
--emit_telemetry BOOLEAN Whether to emit anonymous Safe Synthesizer
telemetry events.
--help Show this message and exit.
Commands:
generate Run the generation stage only.
train Run the training stage only.
```
### Subcommands
- `safe-synthesizer run train` - Run only the training stage, saving the adapter to the run directory.
- `safe-synthesizer run generate` - Run only the generation stage using a saved adapter.
```bash
> uv run safe-synthesizer run generate --help
Usage: safe-synthesizer run generate [OPTIONS]
Run the generation stage only.
This command loads a trained adapter and generates synthetic data. Requires
'run train' to have been executed first.
Use --run-path to specify the exact run directory containing the trained
model, or use --auto-discover-adapter with --artifact-path to automatically
find the latest trained run.
Options:
--config TEXT path to a yaml config file
--data-source TEXT Dataset name, URL, or path to CSV dataset.
[required]
--artifact-path DIRECTORY Base directory for all runs. Runs are
created as <artifact-path>/<config>-
<dataset>/<timestamp>/. [default: ./safe-
synthesizer-artifacts]
--run-path DIRECTORY Explicit path for this run's output
directory. When specified, outputs go
directly to this path. Overrides --artifact-
path.
--output-file PATH Path to output CSV file. Overrides the
default workdir output location.
--log-format [json|plain] Log format for console output. File logging
will always be JSON.
--log-color / --no-log-color Whether to colorize the log output on the
console
--log-file PATH Path to log file. Defaults to a file nested
under the run directory.
-v Verbose logging. 'v' shows debug info from
main program, 'vv' shows debug from
dependencies too
--wandb-mode [online|offline|disabled]
Wandb mode. 'online' will upload logs to
wandb, 'offline' will save logs to a local
file, 'disabled' will not upload logs to
wandb.
--wandb-project TEXT Wandb project. If not specified, the project
will be taken from the environment variable
WANDB_PROJECT.
--auto-discover-adapter Automatically find the latest trained
adapter in --artifact-path. Without this
flag, --run-path must point to a specific
trained run.
--help Show this message and exit.
```
## Managing Configurations
The `config` command provides tools to validate and modify configuration files:
```bash
> uv run safe-synthesizer config --help
Usage: safe-synthesizer config [OPTIONS] COMMAND [ARGS]...
Manage Safe Synthesizer configurations.
Options:
--help Show this message and exit.
Commands:
modify Modify a Safe Synthesizer configuration.
validate Validate a Safe Synthesizer configuration.
```
## Attention Configuration
Safe Synthesizer exposes attention implementation settings for both training and generation.
### Training (`attn_implementation`)
Controls the HuggingFace attention backend used during model loading for training. Set via config YAML, CLI, or SDK:
```yaml
# config.yaml
training:
attn_implementation: "sdpa"
```
```bash
# CLI override
safe-synthesizer run --training__attn_implementation sdpa --data-source my_data.csv
```
| Value | Description | Requires |
|-------|-------------|----------|
| `sdpa` | PyTorch scaled dot product attention (default) | None (built-in) |
| `eager` | Standard PyTorch attention | None (built-in) |
| `kernels-community/flash-attn2` | Flash Attention 2 via HuggingFace Kernels Hub | `kernels` pip package |
| `kernels-community/vllm-flash-attn3` | Flash Attention 3 via HuggingFace Kernels Hub | `kernels` pip package and compatible prebuilt kernel |
| `flash_attention_2` | Flash Attention 2 (traditional) | `flash-attn` pip package |
| `flash_attention_3` | Flash Attention 3 (traditional) | `flash-attn-3` support |
If a `kernels-community/...` value is configured but the `kernels` package is not installed, the backend automatically falls back to `sdpa`.
### Generation (`attention_backend`)
Controls the vLLM attention backend used during synthetic data generation. Defaults to `"auto"`, which lets vLLM auto-select the best available backend.
```yaml
# config.yaml
generation:
attention_backend: "FLASH_ATTN"
```
Common values: `FLASHINFER`, `FLASH_ATTN`, `TORCH_SDPA`, `TRITON_ATTN`, `FLEX_ATTENTION`.
## NIM Integration
Column classification uses a NIM/OpenAI-compatible endpoint to detect entity types
in your data. `NSS_INFERENCE_ENDPOINT` defaults to `https://integrate.api.nvidia.com/v1`;
override it to use a different endpoint.
When using the CLI or Python SDK, set `NSS_INFERENCE_KEY` (and `NSS_INFERENCE_ENDPOINT` only if not
using the default) so column classification can run.
### Local Endpoint
To point to a locally hosted LLM, add the variables to `.env.local` (git-ignored, auto-loaded by mise):
```bash
# .env.local
NSS_INFERENCE_ENDPOINT=https://your-local-nim-endpoint
NSS_INFERENCE_KEY=your-api-key # pragma: allowlist secret
```
Or export them in your shell:
```bash
export NSS_INFERENCE_ENDPOINT="https://your-local-nim-endpoint"
export NSS_INFERENCE_KEY="your-api-key" # pragma: allowlist secret
```
### Disable Classification
To disable classification entirely:
```yaml
replace_pii:
globals:
classify:
enable_classify: false
```
When classification is disabled, NSS falls back to default entity types.
## Artifacts and Workdirs
Safe Synthesizer uses a structured directory format to manage artifacts (trained models, synthetic data, logs).
### Directory Layout
By default, runs are nested under `--artifact-path` using the project name (`<config>---<dataset>`) and a unique run name.
```text
<artifact-path>/<config>---<dataset>/<run_name>/
├── train/
│ ├── safe-synthesizer-config.json
│ └── adapter/ # trained PEFT adapter
│ ├── adapter_config.json
│ ├── adapter_model.safetensors
│ ├── metadata_v2.json
│ └── dataset_schema.json
├── generate/
│ ├── logs.jsonl # generate-only workflow
│ ├── info.json # generate-only workflow
│ ├── synthetic_data.csv
│ ├── evaluation_report.html
│ └── evaluation_metrics.json # machine-readable metrics
├── dataset/
│ ├── training.csv
│ ├── test.csv
│ ├── validation.csv # when training.validation_ratio > 0
│ └── transformed_training.csv # when PII replacement transforms the data
└── logs/
└── <phase>.jsonl # e.g. end_to_end.jsonl or train.jsonl
```
### Run Names
If not provided with `--run-path`, run names are automatically generated using the current `<timestamp>`.
### Overriding Paths
- Use `--run-path` to specify an explicit directory for the run, bypassing the `<project>/<timestamp>` nesting.
- Use `--output-file` to specify an explicit path for the final synthetic CSV, overriding the default location in the `generate/` directory.
## WandB Logging
Safe Synthesizer supports Weights & Biases (WandB) for experiment tracking.
### Configuration
You can enable WandB logging using CLI options or environment variables:
- `--wandb-mode [online|offline|disabled]`: Set the WandB mode. Default is `disabled`.
- `--wandb-project <name>`: Specify the WandB project name.
- `WANDB_API_KEY`: Ensure your API key is set in your environment.
### Logged Data
The following information is logged to WandB:
- Configuration parameters
- Training metrics (if supported by the backend)
- Generation statistics
- Evaluation results
- Timing information
## Dataset Registry
Safe Synthesizer supports a *dataset registry* to simplify working with a standard set of datasets.
Datasets in the registry may be referenced by name, rather than repeatedly specifying long URLS or file paths on the command line.
Additionally, the registry supports custom config overrides or args that are specific to individual datasets.
### Providing a Dataset Registry
You can supply a dataset registry (YAML file) via either the CLI or an environment variable:
- CLI Option:
`--dataset-registry <path_or_url>`
- Environment Variable:
Set `NSS_DATASET_REGISTRY` to point to your YAML file (path or URL).
If both are provided, the CLI option takes precedence.
### Referencing Datasets
When a dataset registry is provided, you can use dataset names defined in the registry with the `--data-source` argument.
For example:
```bash
safe-synthesizer run --dataset-registry my_registry.yaml --data-source my_dataset
```
This will load the dataset from the url plus apply any overrides for `my_dataset` from the registry YAML.
### Dataset Registry YAML Format
The registry file should conform to the pydantic model defined by `DatasetRegistry` in `cli/datasets.py`. For example,
```yaml
# registry.yaml
base_url: /root/data/location
datasets:
- name: dataset1
url: dataset1.csv
- name: dataset2
url: dataset2.jsonl
overrides:
data:
group_training_examples_by: id
- name: dataset3
url: /absolute/path/to/dataset.csv
- name: dataset4
url: https://myhost.com/path/to/dataset.json
load_args:
keyword: custom_arg_for_data_reader
```
- Minimal requirements for each entry in the `datasets:` list are a `name` and a `url`.
`url` may be a URL or a file path, anything that data readers like `pd.read_csv` will accept.
- `base_url` - Any relative urls or paths will be prepended with the `base_url` before attempting to load the dataset.
This only applies to the named datasets in the registry which have a relative url.
Passing a relative `--data-source` on the CLI will attempt to load the file relative to your current working directory, regardless of whether a registry is provided or whether `base_url` is set.
`base_url` is optional, if not provided, it is recommended to use absolute urls or file paths for all entries.
- `overrides` - Dataset specific config overrides, such as a dataset that should always be run with `group_training_examples_by`.
Config values passed as CLI arguments always take precendence, then any overrides from the registry, and finally values from the `--config` yaml file.
- `load_args` - Extra arguments needed by the data reader for a specific dataset.
For example, changing the separator used by `pd.read_csv` for a `.csv` file with a different delimiter.
## Telemetry & Privacy
NeMo Safe Synthesizer includes an optional function to share anonymous telemetry data with NVIDIA for product improvement. Data collected is limited to run-level operational metrics (such as final run status, processing time, record and token counts, configuration parameters, top-level quality and privacy scores, base model used, deployment type, and GPU type). No user or device information is collected. This data is used to prioritize product improvements and will be shared in aggregate with the community. It is not used to track any individual user behavior.
You may opt out of telemetry collection at any time. Opting out applies only to data collection by the NeMo Safe Synthesizer library itself. To disable telemetry in a YAML config, set:
```yaml
emit_telemetry: false
```
To disable telemetry for one CLI invocation, pass `--emit_telemetry false`:
```bash
safe-synthesizer run --emit_telemetry false --data-source my_data.csv
```
To disable telemetry for the current shell, set `NEMO_TELEMETRY_ENABLED=false` (other accepted disabling values: `0`, `no`) in your environment before running:
```bash
export NEMO_TELEMETRY_ENABLED=false
```
Use of third-party endpoints, including NVIDIA Build: NeMo Safe Synthesizer can be configured to use various inference endpoints, including build.nvidia.com (NVIDIA Build). If you choose to use NVIDIA Build or any other third-party endpoint, that endpoint's own terms of service and privacy practices apply independently of this library. Any opt-out you exercise within NeMo Safe Synthesizer does not extend to data collection by your chosen endpoint. NVIDIA Build is intended for evaluation and testing purposes only and may not be used in production environments. Do not submit any confidential information or personal data when using NVIDIA Build.
## License
NeMo Safe Synthesizer is licensed under the [Apache License 2.0](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/blob/main/LICENSE).
## Contact
- [Need help? Ask us a question](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/discussions)
- [Report a bug](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/issues/new?template=bug-report.yml)
- [Make a feature request](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/issues/new?template=feature-request.yml)
- [Report a security vulnerability](https://github.com/NVIDIA-NeMo/Safe-Synthesizer/security/policy)