Python SDK for Datature Vi

The official Python SDK for Datature Vi - a powerful platform for vision-language model training and inference.

🚀 Installation

pip install vi-sdk

For inference capabilities:

pip install vi-sdk[inference]

📖 Documentation

Visit our comprehensive documentation at vi.developers.datature.com

🎯 Quick Start

import vi

# Initialize client (reads DATATURE_VI_SECRET_KEY / DATATURE_VI_ORGANIZATION_ID
# from the environment by default).
client = vi.Client()

# List datasets — first page only.
response = client.datasets.list()
for dataset in response.items:
    print(f"Dataset: {dataset.name}")

# Or iterate every dataset across all pages.
for dataset in client.datasets.list().all_items():
    print(f"Dataset: {dataset.name}")

✨ Features

• Dataset Management: Create, upload, and manage datasets
• Asset Operations: Batch upload and download with progress tracking
• Annotations: JSONL format support for captions and grounding
• Model Training: Track runs and manage trained models
• Inference: Local VLM inference (Qwen2.5-VL, Qwen3-VL, Qwen 3.5/3.6, Gemma 4, NVILA, DeepSeekOCR, plus loader-compatible models — Cosmos Reason1/Reason2, InternVL 3.5, LLaVA-NeXT)
• Dataset Loaders: PyTorch-compatible data loading utilities

📝 License

Apache License 2.0 - see LICENSE file for details.

🤝 Contributing

See CONTRIBUTING.md for contribution guidelines.

Development Setup

We use pre-commit hooks to ensure code quality:

# Install dependencies
uv sync --all-extras

# Install pre-commit hooks
uv run pre-commit install

# Verify setup
uv run pre-commit run --all-files

📖 Documentation:

• Development Guide - Complete development workflow
• Pre-commit Setup - Quick reference for hooks
• Setup Summary - Overview of CI/CD improvements

🐛 Issues

Report issues on GitHub Issues