Four-Way SDK Approach

Weightslab is organized around four complementary capabilities:

  1. Model interaction

  2. Data exploration

  3. Hyperparameter management

  4. Logger and signal tracking

These capabilities are designed to be used together in one training script.

        flowchart LR
  HP[Hyperparameters] --> M[Model Interaction]
  D[Data Exploration] --> M
  M --> L[Logger and Signals]
  L --> D
  L --> HP

Typical integration flow

  • Register hyperparameters first so all components use one shared configuration source.

  • Wrap dataset/dataloader to expose sample IDs and per-sample operations.

  • Wrap model, optimizer, losses, and metrics.

  • Start Weightslab services and run training.

  • Use tags/discards/signals to iteratively improve data and model behavior.