Weightslab Documentation

Weightslab is a Python SDK to inspect, monitor, and edit training behavior for computer vision workflows.

Inspect, edit, and optimize model training with a unified workflow.

Quickstart

Install, build, and run Weightslab documentation locally in minutes.

Quickstart
Four-Way Approach

Understand how model, data, hyperparameters, and logger workflows connect.

Four-Way SDK Approach
Model + Data Control

Learn how to wrap training components and iterate on difficult samples.

Model Interaction
User Functions

Reference all public SDK functions with usage-oriented explanations.

User Functions Reference
User Commands

The weightslab CLI and its interactive console — every command, flag, and default.

User Commands Reference
Examples

Classification, detection, segmentation, clustering, anomaly detection, LiDAR, and Lightning — all with WeightsLab wired in.

Examples
PyTorch Lightning

Integrate Weightslab with Lightning.

PyTorch Lightning Integration
UltraLytics

Integrate Weightslab with Ultralytics.

Ultralytics Integration
Weights Studio

Deploy and operate the UI: architecture, ports, TLS, and actions.

Weights Studio Guide
Configuration

All environment variables for WeightsLab and Weights Studio with defaults and explanations.

Configuration
AI Agent

Drive UI actions (sort, dump, load), data analysis, tagging/discarding, and model freeze/reset with natural language.

AI Agent
gRPC Communication

All RPC handlers, parameters, and behavior. Comprehensive audit logging for user interactions.

gRPC

Weightslab in one sentence

Wrap your training script once, then monitor, tag/discard, adjust, and improve continuously.