graph LR
Model_Persistence["Model Persistence"]
Configuration_Manager["Configuration Manager"]
Diffusion_Process["Diffusion Process"]
Training_Module["Training Module"]
Sampling_Generation_Module["Sampling/Generation Module"]
Model_Persistence -- "Loads Configuration From" --> Configuration_Manager
Model_Persistence -- "Loads Model Into" --> Diffusion_Process
Training_Module -- "Utilizes" --> Model_Persistence
Sampling_Generation_Module -- "Utilizes" --> Model_Persistence
click Model_Persistence href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/genie/Model_Persistence.md" "Details"
The Model Persistence component is crucial for a Deep Learning Research Framework, managing the lifecycle of trained models for experimentation, reproducibility, resuming training, inference, deployment, and version control.
Model Persistence [Expand]
Manages the serialization and deserialization of model checkpoints and associated configurations. It provides utilities to save trained models and load pre-trained models for various tasks like inference, evaluation, or resuming training. It also handles versioning and epoch tracking of saved models.
Related Classes/Methods:
genie.utils.model_io(1:2)genie.utils.model_io:get_versions(9:14)genie.utils.model_io:get_epochs(16:21)
Handles the structured loading and saving of model configurations (hyperparameters, model architecture details).
Related Classes/Methods: None
The core of the model, responsible for tasks like sampling or training after a model is loaded.
Related Classes/Methods:
Periodically saves model checkpoints during training and relies on Model Persistence for saving operations.
Related Classes/Methods: None
Loads pre-trained models for generating new protein structures or performing inference.
Related Classes/Methods: None