Mini-ImagePipeLean CUDA DAG Image Pipeline

A stripped-back GPU runtime focused on DAG execution, multi-stream scheduling, and reusable memory management.

DAG-Native Execution

Build pipelines as dependency graphs and execute them with deterministic topological ordering.

Independent tasks can run across CUDA streams while dependency edges stay synchronized with events.

Pinned and device allocations are pooled so repeated pipeline runs avoid unnecessary churn.

Keep the runtime centered on the built-in operators that define the project’s core value.

Mini-ImagePipe is intentionally centered on four layers: