Concept icon
Concept

Within-Example Parallelization Bottleneck in Recurrent Models

Due to their inherently sequential computation, recurrent models cannot parallelize processing across sequence positions within an individual training example. While parallel processing across distinct examples via batching can partially compensate at shorter lengths, batching becomes severely constrained by device memory limits when dealing with longer sequences, rendering the sequential bottleneck especially critical.

0

1

Concept icon
Updated 2026-09-07

Tags

Prep Sessions

Transformer Architecture and Large Language Model Capabilities @ University of Michigan - Ann Arbor

Ch.1 Transformer Architecture Fundamentals - Transformer Architecture and Large Language Model Capabilities @ University of Michigan - Ann Arbor

Sequential Computation Constraints in Recurrent Networks - Transformer Architecture and Large Language Model Capabilities @ University of Michigan - Ann Arbor