Convert a single PyTorch tensor to TVM Tensor with faster DLPack converter.
(self, tensor: Any)
| 371 | return self._convert_single_pytorch_to_tvm(tensors) |
| 372 | |
| 373 | def _convert_single_pytorch_to_tvm(self, tensor: Any) -> Tensor: |
| 374 | """Convert a single PyTorch tensor to TVM Tensor with faster DLPack converter.""" |
| 375 | # pylint: disable=import-outside-toplevel |
| 376 | import torch |
| 377 | |
| 378 | if isinstance(tensor, Tensor): |
| 379 | return tensor |
| 380 | if isinstance(tensor, torch.Tensor): |
| 381 | # 1. Try faster C++ DLPack converter |
| 382 | if _FASTER_DLPACK_EXTENSION is not None: |
| 383 | try: |
| 384 | dlpack = torch.to_dlpack(tensor) |
| 385 | return tvm.runtime.from_dlpack(dlpack) |
| 386 | except (AttributeError, ValueError): |
| 387 | pass # Fall through to the next method |
| 388 | |
| 389 | # 2. Try modern `torch.to_dlpack` (preferred for PyTorch >= 1.7) |
| 390 | try: |
| 391 | dlpack = torch.to_dlpack(tensor) |
| 392 | return tvm.runtime.from_dlpack(dlpack) |
| 393 | except (AttributeError, ValueError): |
| 394 | pass # Fall through to the next method |
| 395 | |
| 396 | # 3. Try legacy `torch.utils.dlpack.to_dlpack` |
| 397 | if to_dlpack_legacy: |
| 398 | try: |
| 399 | dlpack = to_dlpack_legacy(tensor) |
| 400 | return tvm.runtime.from_dlpack(dlpack) |
| 401 | except (AttributeError, ValueError) as error_legacy: |
| 402 | print( |
| 403 | f"Warning: Legacy DLPack conversion failed ({error_legacy}), " |
| 404 | f"using numpy fallback." |
| 405 | ) |
| 406 | |
| 407 | # 4. If all DLPack methods fail, use numpy fallback |
| 408 | numpy_array = tensor.detach().cpu().numpy() |
| 409 | return tvm.runtime.tensor(numpy_array, device=self.device) |
| 410 | |
| 411 | # For other types (like scalars, lists), convert to numpy first |
| 412 | try: |
| 413 | numpy_array = np.array(tensor, dtype=np.float32) |
| 414 | return tvm.runtime.tensor(numpy_array, device=self.device) |
| 415 | except (TypeError, ValueError) as error: |
| 416 | raise TypeError( |
| 417 | f"Unsupported type for conversion to TVM Tensor: {type(tensor)}" |
| 418 | ) from error |
| 419 | |
| 420 | def _convert_tvm_to_pytorch( |
| 421 | self, tvm_tensors: Any | list[Any] |
no test coverage detected