kernelfoundry.eval_pipeline.tasks.test_custom_task

Celery task for testing and profiling kernel implementations.

Runs correctness tests, performance benchmarks, and profiler traces on custom and reference kernel implementations, optionally inside a container.

kernelfoundry.eval_pipeline.tasks.test_custom_task.test_custom_task(task: dict) → dict[str, dict]

Test and profile a kernel implementation.

Celery task entry point. Decodes the task, runs correctness tests, performance benchmarks, and profiler traces (optionally inside a container), and returns the encoded test results dict.

Parameters:

task – Encoded Task dict as produced by Task.encode().

Returns:

Dict mapping variant keys (e.g., ‘custom’, ‘reference’) to test result dicts (one per variant, encoded as dicts with correctness, performance, and trace results).