Avoid graph break by removing another redundant requires grad false (deepspeedai#7263)

This PR is an follow-up to [PR
deepspeedai#7158](deepspeedai#7158) handling the
same issue in another place.
See [PR deepspeedai#7158](deepspeedai#7158) for
details.

---------

Signed-off-by: Max Kovalenko <[email protected]>
Co-authored-by: Logan Adams <[email protected]>
Co-authored-by: Hongwei Chen <[email protected]>
Signed-off-by: Max Kovalenko <[email protected]>

Author: 3 people

deepspeed/runtime/zero/partition_parameters.py

@@ -1921,7 +1921,6 @@ def _allgather_params(self, param_list, hierarchy=0):
             flat_scale_tensor = torch.empty(scale_tensor_size,
                                             dtype=param_list[0].ds_tensor.ds_quant_scale.dtype,
                                             device=self.local_device)
-            flat_scale_tensor.requires_grad = False
             scale_partitions = []
             for i in range(self.world_size):
                 start = scale_tensor_size * i