Vi SDK
The Vi SDK is a Python library for interacting with Datature Vi programmatically. You can manage datasets, upload assets, and create annotations for phrase grounding, visual question answering (VQA), and freeform text, then start training runs and run inference, all from Python code.
If you prefer working in the browser, the Datature Vi platform covers the same workflows through the UI.
The Vi SDK works with models trained on the Datature Vi platform. Follow the quickstart to train your first model, or install the SDK if you already have one.
- Python 3.10 or later
- A secret key for API authentication
- Your organization ID from your Datature Vi account settings
Child pages hold the depth: one page per REST API resource, a full inference track (model loading, task types, schemas, performance, troubleshooting), and NVIDIA NIM deployment guides. Use the table and cards below to see the full scope, then open the hub that matches your task.
What the SDK covers
Dataset management
Create, list, and download datasets. Supports phrase grounding, visual question answering (VQA), and freeform text formats.
Asset operations
Upload and download images with concurrent transfers, progress tracking, and automatic format detection.
Annotation workflows
Create and retrieve annotations for phrase grounding, VQA, and freeform text tasks.
Model training
Start and monitor training runs on Datature Vi infrastructure.
Inference
Run predictions locally with Qwen3.5, Qwen3-VL, Qwen2.5-VL, NVILA-Lite, Cosmos-Reason1, Cosmos-Reason2, and InternVL3.5.
Type-safe API
Complete type hints, structured error codes, and built-in pagination across all resources.
Platform REST APIs
Guides and reference hubs
Pick a hub when you already know the task. Each hub links to focused pages (for example, inference covers load models, run inference, task types, prediction schemas, performance tuning, and troubleshooting).
Installation
Python environment, optional GPU extras, and a verified install.
Getting started
Authenticate, list datasets, and run a first call from Python.
API resources
All REST-backed resources: datasets through organizations, with examples per resource.
Inference
Load checkpoints, configure generation, task types, schemas, batch and video, performance, troubleshooting.
NIM
Container build, NIM configuration, remote inference, and deployment troubleshooting.
Changelog
Release notes and breaking changes for the Vi SDK package.
Updated 4 days ago