Returns a list of tensors with the all-reduce product across `tensors`. The computation is done with an all-reduce operation, so if only some of the returned tensors are evaluated then the computation will hang. Args: tensors: The input tensors across which to multiply; must be assigned
(tensors)
| 77 | |
| 78 | |
| 79 | def all_prod(tensors): |
| 80 | """Returns a list of tensors with the all-reduce product across `tensors`. |
| 81 | |
| 82 | The computation is done with an all-reduce operation, so if only some of the |
| 83 | returned tensors are evaluated then the computation will hang. |
| 84 | |
| 85 | Args: |
| 86 | tensors: The input tensors across which to multiply; must be assigned |
| 87 | to GPU devices. |
| 88 | |
| 89 | Returns: |
| 90 | List of tensors, each with the product of the input tensors, where tensor i |
| 91 | has the same device as `tensors[i]`. |
| 92 | """ |
| 93 | return _apply_all_reduce('prod', tensors) |
| 94 | |
| 95 | |
| 96 | def all_min(tensors): |
nothing calls this directly
no test coverage detected