Model API Tutorials

This tutorial guides you through using the RoboOrchardLab ModelMixin API to build, save, and load your PyTorch models in a standardized and reproducible way.

The core idea behind ModelMixin is to tightly couple a model’s architecture (Configuration) with its learned weights (State Dictionary). This practice ensures complete model reproducibility: given a saved model directory, you can restore the exact model instance with a single command.

Key Advantages:

• Reproducibility: Bundles configuration with weights, preventing mismatches or forgotten hyperparameters.

• Easy Integration with huggingface: Uses the safetensors format by default for secure and fast file handling and offers an out-of-the-box hook for frameworks like Hugging Face Accelerate.

Creating, Saving, and Loading Models

Creating, Saving, and Loading Models