Datasets:
| from typing import Optional, Tuple | |
| import torch | |
| import torch.distributed as dist | |
| def solution( | |
| grad_fc1_1_lora_A: torch.Tensor, | |
| grad_fc1_2_lora_A: torch.Tensor, | |
| grad_fc2_lora_B: torch.Tensor, | |
| group: Optional[dist.ProcessGroup] = None, | |
| ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: | |
| group = group or dist.group.WORLD | |
| dist.all_reduce(grad_fc1_1_lora_A, op=dist.ReduceOp.SUM, group=group) | |
| dist.all_reduce(grad_fc1_2_lora_A, op=dist.ReduceOp.SUM, group=group) | |
| dist.all_reduce(grad_fc2_lora_B, op=dist.ReduceOp.SUM, group=group) | |
| return grad_fc1_1_lora_A, grad_fc1_2_lora_A, grad_fc2_lora_B | |