Module forward 

Forward execution strategies for [GraphSession].

Each strategy defines HOW input data flows through a compiled graph:

• Standard — full input at once, with inference caching
• Stream — chunked input through [StreamExecutor], respecting StreamMode
• Batched — rows from a DataStore, batch by batch (memory-bounded)

Structs

Batched

Batched forward: read rows from a DataStore in fixed-size batches. Keeps memory bounded — only one batch is materialized at a time.

Standard

Standard forward: full input at once with inference caching.

Stream

Streaming forward: chunk input and process through StreamExecutor. Each filter’s StreamMode (FixedState/Evolving/Barrier) defines its per-chunk contract.

Traits

ForwardStrategy

How a forward pass feeds data through the compiled graph.