Fix 2.8 issue per sample grad #4930

Job	Run time
pytorch_tutorial_build_worker (3, 15, linux.g5.4xlarge.nvidia.gpu)	22m 26s
pytorch_tutorial_build_worker (6, 15, linux.g5.4xlarge.nvidia.gpu)	40m 55s
pytorch_tutorial_build_worker (5, 15, linux.g5.4xlarge.nvidia.gpu)	23m 28s
pytorch_tutorial_build_worker (4, 15, linux.g5.4xlarge.nvidia.gpu)	34m 32s
pytorch_tutorial_build_worker (8, 15, linux.g5.4xlarge.nvidia.gpu)	26m 6s
pytorch_tutorial_build_worker (7, 15, linux.g5.4xlarge.nvidia.gpu)	23m 0s
pytorch_tutorial_build_worker (2, 15, linux.g5.4xlarge.nvidia.gpu)	36m 58s
pytorch_tutorial_build_worker (13, 15, linux.g5.4xlarge.nvidia.gpu)	16m 51s
pytorch_tutorial_build_worker (11, 15, linux.g5.4xlarge.nvidia.gpu)	15m 15s
pytorch_tutorial_build_worker (12, 15, linux.g5.4xlarge.nvidia.gpu)	15m 40s
pytorch_tutorial_build_worker (15, 15, linux.g5.4xlarge.nvidia.gpu)	20m 16s
pytorch_tutorial_build_worker (14, 15, linux.g5.4xlarge.nvidia.gpu)	29m 44s
pytorch_tutorial_build_worker (10, 15, linux.g5.4xlarge.nvidia.gpu)	14m 49s
pytorch_tutorial_build_worker (1, 15, linux.g5.12xlarge.nvidia.gpu)	16m 29s
pytorch_tutorial_build_worker (9, 15, linux.g5.4xlarge.nvidia.gpu)	20m 26s
pytorch_tutorial_build_manager	-1s
	5h 56m 54s

Provide feedback