Diff Coverage

Total: 8 lines
Missing: 1 line
Coverage: 87%

Source File	Diff Coverage (%)	Missing Lines
hyper_parallel/core/fully_shard/api.py	100%
hyper_parallel/platform/torch/fully_shard/param_group.py	0.0%	1085

hyper_parallel/platform/torch/fully_shard/param_group.py

                existing_grad = hsdp_param.sharded_param.grad

            if existing_grad is not None and not hsdp_param.accumulated_allreduced_grad:
                # Get DTensor's local_tensor
                from hyper_parallel.core.dtensor.dtensor import DTensor
                if isinstance(existing_grad, DTensor):
                    existing_grad_local = existing_grad._local_tensor
                else:
                    existing_grad_local = existing_grad