Dria SDK
Dria SDK is a scalable and versatile toolkit for creating and managing synthetic datasets for AI. With Dria, you can orchestrate multi-step pipelines that pull data from both web and siloed sources, blend them with powerful AI model outputs, and produce high-quality synthetic datasets—no GPU required.
Why Dria?
- Dataset Generation: Easily build synthetic data pipelines using Dria’s flexible APIs.
- Multi-Agent Network: Orchestrate complex tasks and data retrieval using specialized agents for web search and siloed APIs.
- No GPUs Needed: Offload your compute to the network, accelerating your workflows without personal GPU hardware.
- Customizable: Define custom Pydantic schemas to shape the output of your datasets precisely.
- Model-Rich: Use different Large Language Models (LLMs) such as OpenAI, Gemini, Ollama or others to synthesize data.
- Grounding & Diversity: Add real-world context to your synthetic datasets with integrated web and siloed data retrieval.
Installation
Dria SDK is available on PyPI. You can install it with:
pip install dria
It’s recommended to use a virtual environment (e.g., virtualenv or conda) to avoid version conflicts with other packages.
Quick Start
Here’s a minimal example to get you started with Dria:
import asyncio
from dria import Prompt, DatasetGenerator, DriaDataset, Model
from pydantic import BaseModel, Field
# 1. Define schema
class Tweet(BaseModel):
topic: str = Field(..., title="Topic")
tweet: str = Field(..., title="Tweet")
# 2. Create a dataset
dataset = DriaDataset(name="tweet_test", description="A dataset of tweets!", schema=Tweet)
# 3. Prepare instructions
instructions = [
{"topic": "BadBadNotGood"},
{"topic": "Decentralized Synthetic Data"}
]
# 4. Create a Prompt
prompter = Prompt(prompt="Write a tweet about {{topic}}", schema=Tweet)
# 5. Generate data
generator = DatasetGenerator(dataset=dataset)
asyncio.run(
generator.generate(
instructions=instructions,
singletons=prompter,
models=Model.GPT4O
)
)
# Convert to Pandas
df = dataset.to_pandas()
print(df)
Output:
topic tweet
0 BadBadNotGood 🎶 Thrilled to have discovered #BadBadNotGood! ...
1 Decentralized Synthetic Data Exploring the future of #AI with decentralized...
Usage
1. Define Your Dataset Schema
Use Pydantic models to define the structure of your synthetic data. For example:
from pydantic import BaseModel, Field
class Tweet(BaseModel):
topic: str = Field(..., title="Topic")
tweet: str = Field(..., title="Tweet")
2. Create a Dataset
Instantiate a DriaDataset by specifying its name, description, and the Pydantic schema:
from dria import DriaDataset
dataset = DriaDataset(name="tweet_test", description="A dataset of tweets!", schema=Tweet)
3. Write a Prompt
Use Prompt objects to define how to generate data from an instruction. You can reference fields using double-curly braces:
from dria import Prompt
prompter = Prompt(
prompt="Write a tweet about {{topic}}",
schema=Tweet
)
4. Generate Synthetic Data
Create a DatasetGenerator and call generate:
from dria import DatasetGenerator, Model
generator = DatasetGenerator(dataset=dataset)
instructions = [{"topic": "Cats"}, {"topic": "Dogs"}]
await generator.generate(
instructions=instructions,
singletons=prompter,
models=Model.GPT4O # Example model
)
instructions: A list of dictionaries, each used to fill the placeholders in yourPrompt.singletons: A single prompt (or list of prompts) that is applied to all instructions.models: The model or list of models you want to use.
5. Convert to Pandas
Finally, convert your generated dataset to a Pandas DataFrame:
import pandas as pd
df = dataset.to_pandas()
print(df)
Dria supports a wide range of data type exports. You can see the full list here. You will need to have some tokens in your balance, which will be approved automatically if required by the register command.
Advanced Usage
Available Models
Dria supports a wide range of models from OpenAI, Gemini, Ollama, and more. You can see the full list here.
Writing Workflows and Custom Pipelines
Dria allows you to write custom workflows and pipelines using the Workflow class. You can see an example of this here.
Structured Outputs
Dria allows you to define custom schemas for your outputs using Pydantic. This allows you to generate highly structured data that can be used for a wide range of applications.
You can see an example of this here.
Parallelization & Offloading
Because Dria tasks can be dispatched to a distributed network of agents, you can leverage massive parallelization without owning any GPUs. This is especially helpful for large-scale data generation tasks:
- Avoid timeouts or rate limits by distributing tasks.
- Scale to thousands or millions of records quickly.
Contributing
Contributions are more than welcome! To get started:
- Fork the repository on GitHub.
- Clone your fork locally and create a new branch for your feature or fix.
- Install dependencies in a virtual environment:
poetry install - Make Changes and Test them thoroughly.
- Submit a Pull Request describing your changes.
We value all contributions—from bug reports and suggestions to feature implementations.
License
This project is licensed under the MIT License. Feel free to use it in your personal or commercial projects.
1 maintainer
Metadata
1 maintainer
Release files for dria 0.0.124
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| dria-0.0.124.tar.gz | 77.8 kB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| dria-0.0.124-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 192.9 kB
Release files / dria-0.0.124.tar.gz
| Download URL | dria-0.0.124.tar.gz |
|---|---|
| Size | 77.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
369776492c1fffacc025e7ffc94e873932e6bc83d3b3cd38fc37128d4fca66eb
|
|
BLAKE2b-256 checksum How to use checksums |
72b3c9f3c613d75e903d4727beb192f44446d014e658e356c3933364f4c30303
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
poetry/1.8.4 CPython/3.10.15 Darwin/23.5.0
|
Release files / dria-0.0.124-py3-none-any.whl
| Download URL | dria-0.0.124-py3-none-any.whl |
|---|---|
| Size | 115.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
f338cf5b27968bf51c12fd1fc4db27a24d6aade3b37c3b464026c3ebbe032cb9
|
|
BLAKE2b-256 checksum How to use checksums |
236dca57ba0ee02a7344ddf8918c350e729baa658af66d2d5c73e4d6b9806a8c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
poetry/1.8.4 CPython/3.10.15 Darwin/23.5.0
|
Security for dria 0.0.124
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
