Aggregated stats for one kernel (name + shape combo).
| 503 | |
| 504 | @dataclass |
| 505 | class KernelRecord: |
| 506 | """Aggregated stats for one kernel (name + shape combo).""" |
| 507 | name: str |
| 508 | op_type: str |
| 509 | gpu_time_us: float |
| 510 | call_count: int |
| 511 | input_shapes: str # string representation of shapes |
| 512 | roofline: str = "" |
| 513 | supported: bool = False |
| 514 | |
| 515 | |
| 516 | def _run_forward(model: nn.Module, inputs: Dict[str, Any]) -> None: |