()
| 142 | |
| 143 | |
| 144 | def _get_module(): |
| 145 | global _module |
| 146 | if _module is None: |
| 147 | _module = compile_cuda(CUDA_SRC, "cross_entropy_cuda") |
| 148 | return _module |
| 149 | |
| 150 | |
| 151 | def kernel_fn(logits: torch.Tensor, targets: torch.Tensor) -> torch.Tensor: |
no test coverage detected