Aaron Gokaslan 3b4b23ab0b [BE][Ez]: Remove extra copy in dtensor parallel loss (#148096) ... Remove an extra copy of the input to `_log_softmax` when there is a dtype and memory format change. Fuse the copies instead. Pull Request resolved: https://github.com/pytorch/pytorch/pull/148096 Approved by: https://github.com/jansel, https://github.com/wconstab		2025-02-28 05:42:32 +00:00
..
__init__.py	[BE][Easy] enable UFMT for torch/distributed/{tensor,_tensor}/ (#128868)	2024-06-18 21:49:02 +00:00
_data_parallel_utils.py	Migrate from Tuple -> tuple in torch/distributed (#144258)	2025-01-10 08:34:54 +00:00
_utils.py	Migrate from Tuple -> tuple in torch/distributed (#144258)	2025-01-10 08:34:54 +00:00
api.py	PEP585 update - torch/distributed/tensor (#145141)	2025-01-18 20:01:59 +00:00
ddp.py	PEP585 update - torch/distributed/tensor (#145141)	2025-01-18 20:01:59 +00:00
fsdp.py	PEP585 update - torch/distributed/tensor (#145141)	2025-01-18 20:01:59 +00:00
input_reshard.py	Migrate from Tuple -> tuple in torch/distributed (#144258)	2025-01-10 08:34:54 +00:00
loss.py	[BE][Ez]: Remove extra copy in dtensor parallel loss (#148096)	2025-02-28 05:42:32 +00:00
style.py	PEP585 update - torch/distributed/tensor (#145141)	2025-01-18 20:01:59 +00:00