Kernel tuning

Inspects, calibrates, and updates processor-aware kernel tuning values.

onnx_light.kernel_tuning.apply_kernel_tuning_updates(*, kernels: Sequence[str] | None = None, element_types: Sequence[int] | None = None, library: str = 'onnx_light', implementation: str | None = None, path: str | None = None, maximum_duration_ms: int = 0, maximum_memory_bytes: int = 0) → dict[str, Any]

Calibrates and persists every calibratable key missing from the cache.

Returns:

The initial proposal, calibration reports, and remaining coverage gaps.

onnx_light.kernel_tuning.propose_kernel_tuning_updates(*, kernels: Sequence[str] | None = None, element_types: Sequence[int] | None = None, library: str = 'onnx_light', implementation: str | None = None, path: str | None = None) → dict[str, Any]

Proposes calibrations for exact kernel keys missing from the local cache.

Parameters:

• kernels – Optional kernel-name subset.

• element_types – Optional ONNX element-type subset.

• library – Tuning library identifier.

• implementation – Optional implementation identifier.

• path – Optional cache path.

Returns:

A report separating calibratable and unsupported missing keys.