Skip to content

Latest commit

 

History

History
59 lines (33 loc) · 2.54 KB

File metadata and controls

59 lines (33 loc) · 2.54 KB
graph LR
    Core_Building_Blocks["Core Building Blocks"]
    Model_Definitions["Model Definitions"]
    Model_Management["Model Management"]
    Training_Evaluation["Training & Evaluation"]
    Core_Building_Blocks -- "provides components to" --> Model_Definitions
    Model_Definitions -- "uses" --> Core_Building_Blocks
    Model_Definitions -- "provided to" --> Model_Management
    Model_Management -- "aggregates" --> Model_Definitions
    Model_Management -- "provides models to" --> Training_Evaluation
    Training_Evaluation -- "consumes models from" --> Model_Management
Loading

CodeBoardingDemoContact

Details

This subsystem focuses on defining, managing, and utilizing neural network architectures, particularly specialized Perforated Bottleneck (PB) models, within the perforatedai project.

Core Building Blocks

Provides the fundamental layers, operations, and utility functions that serve as reusable building blocks for constructing various neural network architectures. This component ensures modularity and consistency across different model implementations.

Related Classes/Methods:

  • perforatedai.core (1:1)

Model Definitions

Implements the specific neural network architectures, including the specialized Perforated Bottleneck (PB) models. This component contains the concrete architectural blueprints and their forward pass logic.

Related Classes/Methods:

  • perforatedai.pb_models

Model Management

Manages the collection of defined neural network models, including orchestrating custom model builders and potentially handling pre-trained model loading. It acts as a central interface for accessing and organizing different model architectures.

Related Classes/Methods:

  • perforatedai.models (1:1)

Training & Evaluation

Handles the processes for training, validation, and evaluation of the neural network models. This includes defining training loops, optimization strategies, and performance metrics. It consumes the models provided by the Model Management component.

Related Classes/Methods:

  • perforatedai.training (1:1)