benchmark-evaluator

KernelBench orchestration automates scanning, dispatching, and aggregating results via a SubAgent workflow. It coordinates serial, level-by-level task execution and reports results back to the scheduler in real time.

Serial kernel benchmarking is best handled by orchestrating tasks individually through a SubAgent. This evaluator dispatches each task sequentially, validates parameters, and enforces a production-grade workflow for per-task reporting.

A SubAgent workflow coordinates kernel benchmark evaluation by invoking a defined task tool interface for each kernel. The scheduler dispatches the SubAgent to generate, verify, and benchmark per task, then collects the results.

No external dependencies are required to run automated kernel benchmark orchestration. You only need to provide the required inputs and define the output path to start scanning tasks and dispatching the SubAgent workflow.

Per-task benchmark results are saved directly to a specified output path during execution. After all tasks finish, the orchestrator generates a final summary report for the scheduling Agent to review the aggregated benchmark data.

Parameter validation is enforced to maintain a production-grade workflow during kernel benchmark scheduling. It ensures that every SubAgent invocation receives correct inputs before task execution and result reporting begins.