Skip to content

Loom_Dynamic_Batching_Engine_Runtime_.md

Latest commit

 

History

History
35 lines (19 loc) · 2.27 KB

Loom_Dynamic_Batching_Engine_Runtime_.md

File metadata and controls

35 lines (19 loc) · 2.27 KB
Raw
graph LR
    Loom_Runtime_Engine["Loom Runtime Engine"]
    Loom_Operation_Definitions["Loom Operation Definitions"]
    Loom_Runtime_Engine -- "is dependent on" --> Loom_Operation_Definitions
    Loom_Operation_Definitions -- "serves as a configuration/definition layer for" --> Loom_Runtime_Engine
Loading

CodeBoardingDemoContact

Details

The Loom Dynamic Batching Engine (Runtime) subsystem is a critical part of the project, focusing on bridging the gap between variable-sized structured inputs and TensorFlow's static graph model.

Loom Runtime Engine

This is the core runtime engine responsible for dynamically batching variable-sized structured inputs into fixed-size tensors. It orchestrates the execution of operations defined by the Loom Operation Definitions, effectively bridging Fold's dynamic input structures with TensorFlow's static graph execution model. It manages the lifecycle of the Loom, including initialization, network construction, input serialization, and output retrieval.

Related Classes/Methods:

Loom Operation Definitions

This component defines the blueprint for all operations executable by the Loom Runtime Engine. It provides the fundamental operations and their necessary type and shape information (input_type_shapes, output_type_shapes) that the Loom Runtime Engine uses to correctly process and batch data. It acts as a configuration and definition layer for the dynamic batching process.

Related Classes/Methods:

FAQ