APIs

API reference: pe.api package

pe.api.API is responsible for implementing the foundation model APIs. It has the following key methods:

• pe.api.API.random_api(): Randomly generates the synthetic samples for the initial samples of the Private Evolution algorithm.

• pe.api.API.variation_api(): Generates the variations of the given synthetic samples for the initial or the next Private Evolution iteration.

Available APIs

Currently, the following APIs are implemented:

• Images

  • pe.api.StableDiffusion: The APIs of Stable Diffusion (introduced in [1]).

  • pe.api.ImprovedDiffusion: The APIs of the improved diffusion model (introduced in [1]).

  • pe.api.DrawText: The APIs that render text on images (introduced in [3]).

  • pe.api.Avatar: The APIs that generate avatars using the Python Avatars library (introduced in [3]).

  • pe.api.NearestImage: The APIs that utilize a static image dataset (introduced in [3]).

• Text

  • pe.api.LLMAugPE: The APIs for text generation using LLMs (introduced in [2]). When constructing the instance of this API, an LLM instance is required. The LLM instances follow the interface of pe.llm.LLM. Currently, the following LLMs are implemented:

    • pe.llm.OpenAILLM: The LLMs from OpenAI APIs.

    • pe.llm.AzureOpenAILLM: The LLMs from Azure OpenAI APIs.

    • pe.llm.HuggingfaceLLM: The open-source LLMs from Huggingface.

• Tabular

  • pe.api.TabularAPI: The APIs for tabular data generation without any model (introduced in [4]). The random API generates synthetic samples by drawing random numbers, and the variation API generates variations by adding small random noise to the samples.

Adding Your Own APIs

To add your own APIs, you need to create a class that inherits from pe.api.API and implements the pe.api.API.random_api() and pe.api.API.variation_api() methods.

Citations

[1] (1,2)

Differentially Private Synthetic Data via Foundation Model APIs 1: Images (ICLR 2024).

[2]

Differentially Private Synthetic Data via Foundation Model APIs 2: Text (ICML 2024 Spotlight).

[3] (1,2,3)

Differentially Private Synthetic Data via APIs 3: Using Simulators Instead of Foundation Models.

[4]

Differentially Private Synthetic Data via APIs 4: Tabular Data (ICML 2026).